comparison vp56.h @ 12349:f618c62a6ddb libavcodec

VP5/6/8: tweak some arithcoder inlining Always inline the arithmetic coder, except in the case of header-parsing stuff, in which case don't inline it at all to save code size.

12348:97a9ea928ffc

#include "x86/vp56_arith.h"
#endif

#ifndef vp56_rac_get_prob
#define vp56_rac_get_prob vp56_rac_get_prob
static inline int vp56_rac_get_prob(VP56RangeCoder *c, uint8_t prob)
{
    unsigned int code_word = vp56_rac_renorm(c);
    unsigned int low = 1 + (((c->high - 1) * prob) >> 8);
    unsigned int low_shift = low << 8;
    int bit = code_word >= low_shift;

    c->high = low;
    c->code_word = code_word;
    return 0;
}

static 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;

    c->code_word = code_word;
    return bit;
}

// rounding is different than vp56_rac_get, is vp56_rac_get wrong?
static inline int vp8_rac_get(VP56RangeCoder *c)
{
    return vp56_rac_get_prob(c, 128);
}

static inline int vp56_rac_gets(VP56RangeCoder *c, int bits)
{
    int value = 0;

    while (bits--) {
        value = (value << 1) | vp56_rac_get(c);
    }

    return value;
}

static inline 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 inline int vp8_rac_get_sint(VP56RangeCoder *c, int bits)
{
    int v;

    if (!vp8_rac_get(c))
        return 0;

    return v;
}

// P(7)
static inline int vp56_rac_gets_nn(VP56RangeCoder *c, int bits)
{
    int v = vp56_rac_gets(c, 7) << 1;
    return v + !v;
}

static inline int vp8_rac_get_nn(VP56RangeCoder *c)
{
    int v = vp8_rac_get_uint(c, 7) << 1;
    return v + !v;
}

static 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]))
            tree += tree->val;
        else

/**
 * 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 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 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 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++);

12349:f618c62a6ddb

#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 code_word = vp56_rac_renorm(c);
    unsigned int low = 1 + (((c->high - 1) * prob) >> 8);
    unsigned int low_shift = low << 8;
    int bit = code_word >= low_shift;

    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;

    c->code_word = code_word;
    return bit;
}

// 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 int vp56_rac_gets(VP56RangeCoder *c, int bits)
{
    int value = 0;

    while (bits--) {
        value = (value << 1) | vp56_rac_get(c);
    }

    return value;
}

static 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 int vp8_rac_get_sint(VP56RangeCoder *c, int bits)
{
    int v;

    if (!vp8_rac_get(c))
        return 0;

    return v;
}

// P(7)
static int vp56_rac_gets_nn(VP56RangeCoder *c, int bits)
{
    int v = vp56_rac_gets(c, 7) << 1;
    return v + !v;
}

static 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]))
            tree += tree->val;
        else

/**
 * 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++);