/**
- * @file libavcodec/vp56.h
+ * @file
* VP5 and VP6 compatible video decoder (common features)
*
* Copyright (C) 2006 Aurelien Jacobs <aurel@gnuage.org>
#include "dsputil.h"
#include "get_bits.h"
#include "bytestream.h"
-
+#include "cabac.h"
+#include "vp56dsp.h"
typedef struct vp56_context VP56Context;
-typedef struct vp56_mv VP56mv;
+
+typedef struct {
+ int16_t x;
+ int16_t y;
+} DECLARE_ALIGNED(4, , VP56mv);
typedef void (*VP56ParseVectorAdjustment)(VP56Context *s,
VP56mv *vect);
-typedef int (*VP56Adjust)(int v, int t);
typedef void (*VP56Filter)(VP56Context *s, uint8_t *dst, uint8_t *src,
int offset1, int offset2, int stride,
VP56mv mv, int mask, int select, int luma);
typedef struct {
int high;
- int bits;
+ int bits; /* stored negated (i.e. negative "bits" is a positive number of
+ bits left) in order to eliminate a negate in cache refilling */
const uint8_t *buffer;
- unsigned long code_word;
+ const uint8_t *end;
+ unsigned int code_word;
} VP56RangeCoder;
typedef struct {
DCTELEM dc_coeff;
} VP56RefDc;
-struct vp56_mv {
- int x;
- int y;
-};
-
typedef struct {
uint8_t type;
VP56mv mv;
struct vp56_context {
AVCodecContext *avctx;
DSPContext dsp;
+ VP56DSPContext vp56dsp;
ScanTable scantable;
AVFrame frames[4];
AVFrame *framep[6];
int quantizer;
uint16_t dequant_dc;
uint16_t dequant_ac;
+ int8_t *qscale_table;
/* DC predictors management */
VP56RefDc *above_blocks;
/* blocks / macroblock */
VP56mb mb_type;
VP56Macroblock *macroblocks;
- DECLARE_ALIGNED_16(DCTELEM, block_coeff[6][64]);
+ DECLARE_ALIGNED(16, DCTELEM, block_coeff)[6][64];
/* motion vectors */
VP56mv mv[6]; /* vectors for each block in MB */
const uint8_t *vp56_coord_div;
VP56ParseVectorAdjustment parse_vector_adjustment;
- VP56Adjust adjust;
VP56Filter filter;
VP56ParseCoeff parse_coeff;
VP56DefaultModelsInit default_models_init;
};
-void vp56_init(AVCodecContext *avctx, int flip, int has_alpha);
-int vp56_free(AVCodecContext *avctx);
-void vp56_init_dequant(VP56Context *s, int quantizer);
-int vp56_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
- AVPacket *avpkt);
+void ff_vp56_init(AVCodecContext *avctx, int flip, int has_alpha);
+int ff_vp56_free(AVCodecContext *avctx);
+void ff_vp56_init_dequant(VP56Context *s, int quantizer);
+int ff_vp56_decode_frame(AVCodecContext *avctx, void *data, int *data_size,
+ AVPacket *avpkt);
/**
* vp56 specific range coder implementation
*/
-static inline void vp56_init_range_decoder(VP56RangeCoder *c,
- const uint8_t *buf, int buf_size)
+extern const uint8_t ff_vp56_norm_shift[256];
+void ff_vp56_init_range_decoder(VP56RangeCoder *c, const uint8_t *buf, int buf_size);
+
+static av_always_inline unsigned int vp56_rac_renorm(VP56RangeCoder *c)
{
- c->high = 255;
- c->bits = 8;
- c->buffer = buf;
- c->code_word = bytestream_get_be16(&c->buffer);
+ int shift = ff_vp56_norm_shift[c->high];
+ int bits = c->bits;
+ unsigned int code_word = c->code_word;
+
+ c->high <<= shift;
+ code_word <<= shift;
+ bits += shift;
+ if(bits >= 0 && c->buffer < c->end) {
+ code_word |= bytestream_get_be16(&c->buffer) << bits;
+ bits -= 16;
+ }
+ c->bits = bits;
+ return code_word;
}
-static inline int vp56_rac_get_prob(VP56RangeCoder *c, uint8_t prob)
+#if ARCH_X86
+#include "x86/vp56_arith.h"
+#endif
+
+#ifndef vp56_rac_get_prob
+#define vp56_rac_get_prob vp56_rac_get_prob
+static av_always_inline int vp56_rac_get_prob(VP56RangeCoder *c, uint8_t prob)
{
- unsigned int low = 1 + (((c->high - 1) * prob) / 256);
- unsigned int low_shift = low << 8;
- int bit = c->code_word >= low_shift;
+ unsigned int code_word = vp56_rac_renorm(c);
+ unsigned int low = 1 + (((c->high - 1) * prob) >> 8);
+ unsigned int low_shift = low << 16;
+ int bit = code_word >= low_shift;
- if (bit) {
- c->high -= low;
- c->code_word -= low_shift;
- } else {
- c->high = low;
- }
+ c->high = bit ? c->high - low : low;
+ c->code_word = bit ? code_word - low_shift : code_word;
- /* normalize */
- while (c->high < 128) {
- c->high <<= 1;
- c->code_word <<= 1;
- if (--c->bits == 0) {
- c->bits = 8;
- c->code_word |= *c->buffer++;
- }
- }
return bit;
}
+#endif
-static inline int vp56_rac_get(VP56RangeCoder *c)
+// branchy variant, to be used where there's a branch based on the bit decoded
+static av_always_inline int vp56_rac_get_prob_branchy(VP56RangeCoder *c, int prob)
{
+ unsigned long code_word = vp56_rac_renorm(c);
+ unsigned low = 1 + (((c->high - 1) * prob) >> 8);
+ unsigned low_shift = low << 16;
+
+ if (code_word >= low_shift) {
+ c->high -= low;
+ c->code_word = code_word - low_shift;
+ return 1;
+ }
+
+ c->high = low;
+ c->code_word = code_word;
+ return 0;
+}
+
+static av_always_inline int vp56_rac_get(VP56RangeCoder *c)
+{
+ unsigned int code_word = vp56_rac_renorm(c);
/* equiprobable */
int low = (c->high + 1) >> 1;
- unsigned int low_shift = low << 8;
- int bit = c->code_word >= low_shift;
+ unsigned int low_shift = low << 16;
+ int bit = code_word >= low_shift;
if (bit) {
- c->high = (c->high - low) << 1;
- c->code_word -= low_shift;
+ c->high -= low;
+ code_word -= low_shift;
} else {
- c->high = low << 1;
+ c->high = low;
}
- /* normalize */
- c->code_word <<= 1;
- if (--c->bits == 0) {
- c->bits = 8;
- c->code_word |= *c->buffer++;
- }
+ c->code_word = code_word;
return bit;
}
-static inline int vp56_rac_gets(VP56RangeCoder *c, int bits)
+// rounding is different than vp56_rac_get, is vp56_rac_get wrong?
+static av_always_inline int vp8_rac_get(VP56RangeCoder *c)
+{
+ return vp56_rac_get_prob(c, 128);
+}
+
+static av_unused int vp56_rac_gets(VP56RangeCoder *c, int bits)
{
int value = 0;
return value;
}
-static inline int vp56_rac_gets_nn(VP56RangeCoder *c, int bits)
+static av_unused int vp8_rac_get_uint(VP56RangeCoder *c, int bits)
+{
+ int value = 0;
+
+ while (bits--) {
+ value = (value << 1) | vp8_rac_get(c);
+ }
+
+ return value;
+}
+
+// fixme: add 1 bit to all the calls to this?
+static av_unused int vp8_rac_get_sint(VP56RangeCoder *c, int bits)
+{
+ int v;
+
+ if (!vp8_rac_get(c))
+ return 0;
+
+ v = vp8_rac_get_uint(c, bits);
+
+ if (vp8_rac_get(c))
+ v = -v;
+
+ return v;
+}
+
+// P(7)
+static av_unused int vp56_rac_gets_nn(VP56RangeCoder *c, int bits)
{
int v = vp56_rac_gets(c, 7) << 1;
return v + !v;
}
-static inline int vp56_rac_get_tree(VP56RangeCoder *c,
- const VP56Tree *tree,
- const uint8_t *probs)
+static av_unused int vp8_rac_get_nn(VP56RangeCoder *c)
+{
+ int v = vp8_rac_get_uint(c, 7) << 1;
+ return v + !v;
+}
+
+static av_always_inline
+int vp56_rac_get_tree(VP56RangeCoder *c,
+ const VP56Tree *tree,
+ const uint8_t *probs)
{
while (tree->val > 0) {
if (vp56_rac_get_prob(c, probs[tree->prob_idx]))
return -tree->val;
}
+/**
+ * This is identical to vp8_rac_get_tree except for the possibility of starting
+ * on a node other than the root node, needed for coeff decode where this is
+ * used to save a bit after a 0 token (by disallowing EOB to immediately follow.)
+ */
+static av_always_inline
+int vp8_rac_get_tree_with_offset(VP56RangeCoder *c, const int8_t (*tree)[2],
+ const uint8_t *probs, int i)
+{
+ do {
+ i = tree[i][vp56_rac_get_prob(c, probs[i])];
+ } while (i > 0);
+
+ return -i;
+}
+
+// how probabilities are associated with decisions is different I think
+// well, the new scheme fits in the old but this way has one fewer branches per decision
+static av_always_inline
+int vp8_rac_get_tree(VP56RangeCoder *c, const int8_t (*tree)[2],
+ const uint8_t *probs)
+{
+ return vp8_rac_get_tree_with_offset(c, tree, probs, 0);
+}
+
+// DCTextra
+static av_always_inline int vp8_rac_get_coeff(VP56RangeCoder *c, const uint8_t *prob)
+{
+ int v = 0;
+
+ do {
+ v = (v<<1) + vp56_rac_get_prob(c, *prob++);
+ } while (*prob);
+
+ return v;
+}
+
#endif /* AVCODEC_VP56_H */