static void encode_block(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;
struct optimize_ctx *const ctx = args->ctx;
struct macroblock_plane *const p = &x->plane[plane];
struct macroblockd_plane *const pd = &xd->plane[plane];
int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
int i, j;
uint8_t *dst;
txfrm_block_to_raster_xy(plane_bsize, tx_size, block, &i, &j);
dst = &pd->dst.buf[4 * j * pd->dst.stride + 4 * i];
// TODO(jingning): per transformed block zero forcing only enabled for
// luma component. will integrate chroma components as well.
if (x->zcoeff_blk[tx_size][block] && plane == 0) {
p->eobs[block] = 0;
ctx->ta[plane][i] = 0;
ctx->tl[plane][j] = 0;
return;
}
if (!x->skip_recode)
vp9_xform_quant(x, plane, block, plane_bsize, tx_size);
if (x->optimize && (!x->skip_recode || !x->skip_optimize)) {
optimize_b(plane, block, plane_bsize, tx_size, x, ctx);
} else {
ctx->ta[plane][i] = p->eobs[block] > 0;
ctx->tl[plane][j] = p->eobs[block] > 0;
}
if (p->eobs[block])
*(args->skip) = 0;
if (x->skip_encode || p->eobs[block] == 0)
return;
switch (tx_size) {
case TX_32X32:
vp9_idct32x32_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
break;
case TX_16X16:
vp9_idct16x16_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
break;
case TX_8X8:
vp9_idct8x8_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
break;
case TX_4X4:
// 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, pd->dst.stride, p->eobs[block]);
break;
default:
assert(0 && "Invalid transform size");
}
}
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 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[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;
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)];
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(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(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[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;
}
