void vp9_inverse_transform_mby_8x8(MACROBLOCKD *xd) {
int i;
BLOCKD *blockd = xd->block;
int has_2nd_order = get_2nd_order_usage(xd);
if (has_2nd_order) {
// do 2nd order transform on the dc block
vp9_short_ihaar2x2(blockd[24].dqcoeff, blockd[24].diff, 8);
recon_dcblock_8x8(xd); // need to change for 8x8
}
for (i = 0; i < 9; i += 8) {
TX_TYPE tx_type = get_tx_type_8x8(xd, &xd->block[i]);
if (tx_type != DCT_DCT) {
vp9_ihtllm(xd->block[i].dqcoeff, xd->block[i].diff, 32, tx_type, 8,
xd->block[i].eob);
} else {
vp9_inverse_transform_b_8x8(&blockd[i].dqcoeff[0],
&blockd[i].diff[0], 32);
}
}
for (i = 2; i < 11; i += 8) {
TX_TYPE tx_type = get_tx_type_8x8(xd, &xd->block[i]);
if (tx_type != DCT_DCT) {
vp9_ihtllm(xd->block[i + 2].dqcoeff, xd->block[i].diff, 32, tx_type, 8,
xd->block[i + 2].eob);
} else {
vp9_inverse_transform_b_8x8(&blockd[i + 2].dqcoeff[0],
&blockd[i].diff[0], 32);
}
}
}
static void decode_block(int plane, int block, BLOCK_SIZE_TYPE bsize,
int ss_txfrm_size, void *arg) {
MACROBLOCKD* const xd = arg;
struct macroblockd_plane *pd = &xd->plane[plane];
int16_t* const qcoeff = BLOCK_OFFSET(pd->qcoeff, block, 16);
const int stride = pd->dst.stride;
const int raster_block = txfrm_block_to_raster_block(xd, bsize, plane,
block, ss_txfrm_size);
uint8_t* const dst = raster_block_offset_uint8(xd, bsize, plane,
raster_block,
pd->dst.buf, stride);
TX_TYPE tx_type;
switch (ss_txfrm_size / 2) {
case TX_4X4:
tx_type = plane == 0 ? get_tx_type_4x4(xd, raster_block) : DCT_DCT;
if (tx_type == DCT_DCT)
xd->itxm_add(qcoeff, dst, stride, pd->eobs[block]);
else
vp9_iht_add_c(tx_type, qcoeff, dst, stride, pd->eobs[block]);
break;
case TX_8X8:
tx_type = plane == 0 ? get_tx_type_8x8(xd, raster_block) : DCT_DCT;
vp9_iht_add_8x8_c(tx_type, qcoeff, dst, stride, pd->eobs[block]);
break;
case TX_16X16:
tx_type = plane == 0 ? get_tx_type_16x16(xd, raster_block) : DCT_DCT;
vp9_iht_add_16x16_c(tx_type, qcoeff, dst, stride, pd->eobs[block]);
break;
case TX_32X32:
vp9_idct_add_32x32(qcoeff, dst, stride, pd->eobs[block]);
break;
}
}
void vp9_encode_intra8x8(MACROBLOCK *x, int ib) {
MACROBLOCKD *xd = &x->e_mbd;
BLOCKD *b = &xd->block[ib];
BLOCK *be = &x->block[ib];
const int iblock[4] = {0, 1, 4, 5};
int i;
TX_TYPE tx_type;
#if CONFIG_COMP_INTRA_PRED
if (b->bmi.as_mode.second == (MB_PREDICTION_MODE)(DC_PRED - 1)) {
#endif
vp9_intra8x8_predict(b, b->bmi.as_mode.first, b->predictor);
#if CONFIG_COMP_INTRA_PRED
} else {
vp9_comp_intra8x8_predict(b, b->bmi.as_mode.first, b->bmi.as_mode.second,
b->predictor);
}
#endif
// generate residual blocks
vp9_subtract_4b_c(be, b, 16);
if (xd->mode_info_context->mbmi.txfm_size == TX_8X8) {
int idx = (ib & 0x02) ? (ib + 2) : ib;
tx_type = get_tx_type_8x8(xd, &xd->block[ib]);
if (tx_type != DCT_DCT) {
vp9_fht(be->src_diff, 32, (x->block + idx)->coeff,
tx_type, 8);
x->quantize_b_8x8(x->block + idx, xd->block + idx);
vp9_ihtllm(xd->block[idx].dqcoeff, xd->block[ib].diff, 32,
tx_type, 8, xd->block[idx].eob);
} else {
x->vp9_short_fdct8x8(be->src_diff, (x->block + idx)->coeff, 32);
x->quantize_b_8x8(x->block + idx, xd->block + idx);
vp9_short_idct8x8(xd->block[idx].dqcoeff, xd->block[ib].diff, 32);
}
} else {
for (i = 0; i < 4; i++) {
b = &xd->block[ib + iblock[i]];
be = &x->block[ib + iblock[i]];
tx_type = get_tx_type_4x4(xd, b);
if (tx_type != DCT_DCT) {
vp9_fht_c(be->src_diff, 32, be->coeff, tx_type, 4);
vp9_ht_quantize_b_4x4(be, b, tx_type);
vp9_ihtllm(b->dqcoeff, b->diff, 32, tx_type, 4, b->eob);
} else {
x->vp9_short_fdct4x4(be->src_diff, be->coeff, 32);
x->quantize_b_4x4(be, b);
vp9_inverse_transform_b_4x4(xd, ib + iblock[i], 32);
}
}
}
// reconstruct submacroblock
for (i = 0; i < 4; i++) {
b = &xd->block[ib + iblock[i]];
vp9_recon_b_c(b->predictor, b->diff, *(b->base_dst) + b->dst,
b->dst_stride);
}
}
void vp9_transform_mby_8x8(MACROBLOCK *x) {
int i;
MACROBLOCKD *xd = &x->e_mbd;
TX_TYPE tx_type;
int has_2nd_order = get_2nd_order_usage(xd);
for (i = 0; i < 9; i += 8) {
BLOCK *b = &x->block[i];
tx_type = get_tx_type_8x8(xd, &xd->block[i]);
if (tx_type != DCT_DCT) {
assert(has_2nd_order == 0);
vp9_fht_c(b->src_diff, 32, b->coeff, tx_type, 8);
} else {
x->vp9_short_fdct8x8(&x->block[i].src_diff[0],
&x->block[i].coeff[0], 32);
}
}
for (i = 2; i < 11; i += 8) {
BLOCK *b = &x->block[i];
tx_type = get_tx_type_8x8(xd, &xd->block[i]);
if (tx_type != DCT_DCT) {
assert(has_2nd_order == 0);
vp9_fht_c(b->src_diff, 32, (b + 2)->coeff, tx_type, 8);
} else {
x->vp9_short_fdct8x8(&x->block[i].src_diff[0],
&x->block[i + 2].coeff[0], 32);
}
}
if (has_2nd_order) {
// build dc block from 2x2 y dc values
build_dcblock_8x8(x);
// do 2nd order transform on the dc block
x->short_fhaar2x2(&x->block[24].src_diff[0],
&x->block[24].coeff[0], 8);
} else {
vpx_memset(x->block[24].coeff, 0, 16 * sizeof(x->block[24].coeff[0]));
}
}
static void inverse_transform_block(MACROBLOCKD* xd, int plane, int block,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size) {
struct macroblockd_plane *const pd = &xd->plane[plane];
int16_t* const qcoeff = BLOCK_OFFSET(pd->qcoeff, block);
const int stride = pd->dst.stride;
const int eob = pd->eobs[block];
if (eob > 0) {
TX_TYPE tx_type;
const int raster_block = txfrm_block_to_raster_block(plane_bsize, tx_size,
block);
uint8_t* const dst = raster_block_offset_uint8(plane_bsize, raster_block,
pd->dst.buf, stride);
switch (tx_size) {
case TX_4X4:
tx_type = get_tx_type_4x4(pd->plane_type, xd, raster_block);
if (tx_type == DCT_DCT)
xd->itxm_add(qcoeff, dst, stride, eob);
else
vp9_iht4x4_add(tx_type, qcoeff, dst, stride, eob);
break;
case TX_8X8:
tx_type = get_tx_type_8x8(pd->plane_type, xd);
vp9_iht8x8_add(tx_type, qcoeff, dst, stride, eob);
break;
case TX_16X16:
tx_type = get_tx_type_16x16(pd->plane_type, xd);
vp9_iht16x16_add(tx_type, qcoeff, dst, stride, eob);
break;
case TX_32X32:
tx_type = DCT_DCT;
vp9_idct32x32_add(qcoeff, dst, stride, eob);
break;
default:
assert(!"Invalid transform size");
}
if (eob == 1) {
vpx_memset(qcoeff, 0, 2 * sizeof(qcoeff[0]));
} else {
if (tx_type == DCT_DCT && tx_size <= TX_16X16 && eob <= 10)
vpx_memset(qcoeff, 0, 4 * (4 << tx_size) * sizeof(qcoeff[0]));
else
vpx_memset(qcoeff, 0, (16 << (tx_size << 1)) * sizeof(qcoeff[0]));
}
}
}
static void tokenize_b(int plane, int block, BLOCK_SIZE_TYPE plane_bsize,
TX_SIZE tx_size, void *arg) {
struct tokenize_b_args* const args = arg;
VP9_COMP *cpi = args->cpi;
MACROBLOCKD *xd = args->xd;
TOKENEXTRA **tp = args->tp;
struct macroblockd_plane *pd = &xd->plane[plane];
MB_MODE_INFO *mbmi = &xd->mode_info_context->mbmi;
int pt; /* near block/prev token context index */
int c = 0, rc = 0;
TOKENEXTRA *t = *tp; /* store tokens starting here */
const int eob = pd->eobs[block];
const PLANE_TYPE type = pd->plane_type;
const int16_t *qcoeff_ptr = BLOCK_OFFSET(pd->qcoeff, block);
int seg_eob;
const int segment_id = mbmi->segment_id;
const int16_t *scan, *nb;
vp9_coeff_count *const counts = cpi->coef_counts[tx_size];
vp9_coeff_probs_model *const coef_probs = cpi->common.fc.coef_probs[tx_size];
const int ref = is_inter_block(mbmi);
ENTROPY_CONTEXT above_ec, left_ec;
uint8_t token_cache[1024];
const uint8_t *band_translate;
ENTROPY_CONTEXT *A, *L;
int aoff, loff;
txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &aoff, &loff);
A = pd->above_context + aoff;
L = pd->left_context + loff;
assert((!type && !plane) || (type && plane));
switch (tx_size) {
case TX_4X4:
above_ec = A[0] != 0;
left_ec = L[0] != 0;
seg_eob = 16;
scan = get_scan_4x4(get_tx_type_4x4(type, xd, block));
band_translate = vp9_coefband_trans_4x4;
break;
case TX_8X8:
above_ec = !!*(uint16_t *)A;
left_ec = !!*(uint16_t *)L;
seg_eob = 64;
scan = get_scan_8x8(get_tx_type_8x8(type, xd));
band_translate = vp9_coefband_trans_8x8plus;
break;
case TX_16X16:
above_ec = !!*(uint32_t *)A;
left_ec = !!*(uint32_t *)L;
seg_eob = 256;
scan = get_scan_16x16(get_tx_type_16x16(type, xd));
band_translate = vp9_coefband_trans_8x8plus;
break;
case TX_32X32:
above_ec = !!*(uint64_t *)A;
left_ec = !!*(uint64_t *)L;
seg_eob = 1024;
scan = vp9_default_scan_32x32;
band_translate = vp9_coefband_trans_8x8plus;
break;
default:
assert(!"Invalid transform size");
}
pt = combine_entropy_contexts(above_ec, left_ec);
nb = vp9_get_coef_neighbors_handle(scan);
if (vp9_segfeature_active(&cpi->common.seg, segment_id, SEG_LVL_SKIP))
seg_eob = 0;
c = 0;
do {
const int band = get_coef_band(band_translate, c);
int token;
int v = 0;
rc = scan[c];
if (c)
pt = get_coef_context(nb, token_cache, c);
if (c < eob) {
v = qcoeff_ptr[rc];
assert(-DCT_MAX_VALUE <= v && v < DCT_MAX_VALUE);
t->extra = vp9_dct_value_tokens_ptr[v].extra;
token = vp9_dct_value_tokens_ptr[v].token;
} else {
token = DCT_EOB_TOKEN;
}
t->token = token;
t->context_tree = coef_probs[type][ref][band][pt];
t->skip_eob_node = (c > 0) && (token_cache[scan[c - 1]] == 0);
assert(vp9_coef_encodings[t->token].len - t->skip_eob_node > 0);
++counts[type][ref][band][pt][token];
if (!t->skip_eob_node)
++cpi->common.counts.eob_branch[tx_size][type][ref][band][pt];
token_cache[rc] = vp9_pt_energy_class[token];
++t;
} while (c < eob && ++c < seg_eob);
*tp = t;
set_contexts(xd, pd, plane_bsize, tx_size, c > 0, aoff, loff);
}
static void encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {
struct encode_b_args* const args = arg;
MACROBLOCK *const x = args->x;
MACROBLOCKD *const xd = &x->e_mbd;
MB_MODE_INFO *mbmi = &xd->mi_8x8[0]->mbmi;
struct macroblock_plane *const p = &x->plane[plane];
struct macroblockd_plane *const pd = &xd->plane[plane];
int16_t *coeff = BLOCK_OFFSET(p->coeff, block);
int16_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
int16_t *dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
const scan_order *scan_order;
TX_TYPE tx_type;
MB_PREDICTION_MODE mode;
const int bwl = b_width_log2(plane_bsize);
const int diff_stride = 4 * (1 << bwl);
uint8_t *src, *dst;
int16_t *src_diff;
uint16_t *eob = &p->eobs[block];
const int src_stride = p->src.stride;
const int dst_stride = pd->dst.stride;
int i, j;
txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
dst = &pd->dst.buf[4 * (j * dst_stride + i)];
src = &p->src.buf[4 * (j * src_stride + i)];
src_diff = &p->src_diff[4 * (j * diff_stride + i)];
// if (x->optimize)
// optimize_b(plane, block, plane_bsize, tx_size, x, args->ctx);
switch (tx_size) {
case TX_32X32:
scan_order = &vp9_default_scan_orders[TX_32X32];
mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
vp9_predict_intra_block(xd, block >> 6, bwl, TX_32X32, mode,
x->skip_encode ? src : dst,
x->skip_encode ? src_stride : dst_stride,
dst, dst_stride, i, j, plane);
if (!x->skip_recode) {
vp9_subtract_block(32, 32, src_diff, diff_stride,
src, src_stride, dst, dst_stride);
fdct32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride);
vp9_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, p->zbin_extra, eob, scan_order->scan,
scan_order->iscan);
}
if (!x->skip_encode && *eob)
vp9_idct32x32_add(dqcoeff, dst, dst_stride, *eob);
break;
case TX_16X16:
tx_type = get_tx_type_16x16(pd->plane_type, xd);
scan_order = &vp9_scan_orders[TX_16X16][tx_type];
mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
vp9_predict_intra_block(xd, block >> 4, bwl, TX_16X16, mode,
x->skip_encode ? src : dst,
x->skip_encode ? src_stride : dst_stride,
dst, dst_stride, i, j, plane);
if (!x->skip_recode) {
vp9_subtract_block(16, 16, src_diff, diff_stride,
src, src_stride, dst, dst_stride);
vp9_fht16x16(src_diff, coeff, diff_stride, tx_type);
vp9_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, p->zbin_extra, eob, scan_order->scan,
scan_order->iscan);
}
if (!x->skip_encode && *eob)
vp9_iht16x16_add(tx_type, dqcoeff, dst, dst_stride, *eob);
break;
case TX_8X8:
tx_type = get_tx_type_8x8(pd->plane_type, xd);
scan_order = &vp9_scan_orders[TX_8X8][tx_type];
mode = plane == 0 ? mbmi->mode : mbmi->uv_mode;
vp9_predict_intra_block(xd, block >> 2, bwl, TX_8X8, mode,
x->skip_encode ? src : dst,
x->skip_encode ? src_stride : dst_stride,
dst, dst_stride, i, j, plane);
if (!x->skip_recode) {
vp9_subtract_block(8, 8, src_diff, diff_stride,
src, src_stride, dst, dst_stride);
vp9_fht8x8(src_diff, coeff, diff_stride, tx_type);
vp9_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round, p->quant,
p->quant_shift, qcoeff, dqcoeff,
pd->dequant, p->zbin_extra, eob, scan_order->scan,
scan_order->iscan);
}
if (!x->skip_encode && *eob)
vp9_iht8x8_add(tx_type, dqcoeff, dst, dst_stride, *eob);
break;
case TX_4X4:
tx_type = get_tx_type_4x4(pd->plane_type, xd, block);
scan_order = &vp9_scan_orders[TX_4X4][tx_type];
mode = plane == 0 ? get_y_mode(xd->mi_8x8[0], block) : mbmi->uv_mode;
vp9_predict_intra_block(xd, block, bwl, TX_4X4, mode,
x->skip_encode ? src : dst,
x->skip_encode ? src_stride : dst_stride,
dst, dst_stride, i, j, plane);
if (!x->skip_recode) {
vp9_subtract_block(4, 4, src_diff, diff_stride,
src, src_stride, dst, dst_stride);
if (tx_type != DCT_DCT)
vp9_fht4x4(src_diff, coeff, diff_stride, tx_type);
else
x->fwd_txm4x4(src_diff, coeff, diff_stride);
vp9_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round, p->quant,
p->quant_shift, qcoeff, dqcoeff,
pd->dequant, p->zbin_extra, eob, scan_order->scan,
scan_order->iscan);
}
if (!x->skip_encode && *eob) {
if (tx_type == DCT_DCT)
// this is like vp9_short_idct4x4 but has a special case around eob<=1
// which is significant (not just an optimization) for the lossless
// case.
xd->itxm_add(dqcoeff, dst, dst_stride, *eob);
else
vp9_iht4x4_16_add(dqcoeff, dst, dst_stride, tx_type);
}
break;
default:
assert(0);
}
if (*eob)
*(args->skip) = 0;
}
