vpx/vp9/encoder/vp9_quantize.c

278 lines
9.3 KiB
C
Raw Normal View History

2010-05-18 17:58:33 +02:00
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
2010-05-18 17:58:33 +02:00
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
2010-05-18 17:58:33 +02:00
*/
#include <math.h>
#include "vpx_mem/vpx_mem.h"
#include "vp9/encoder/vp9_onyx_int.h"
#include "vp9/encoder/vp9_quantize.h"
#include "vp9/common/vp9_quant_common.h"
2010-05-18 17:58:33 +02:00
#include "vp9/common/vp9_seg_common.h"
2011-10-05 12:26:00 +02:00
#ifdef ENC_DEBUG
extern int enc_debug;
#endif
static INLINE int plane_idx(int plane) {
return plane == 0 ? 0 :
plane == 1 ? 16 : 20;
}
static void quantize(int16_t *zbin_boost_orig_ptr,
int16_t *coeff_ptr, int n_coeffs, int skip_block,
int16_t *zbin_ptr, int16_t *round_ptr, int16_t *quant_ptr,
uint8_t *quant_shift_ptr,
int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr,
int16_t *dequant_ptr, int zbin_oq_value,
uint16_t *eob_ptr,
const int *scan, int mul) {
int i, rc, eob;
int zbin;
int x, y, z, sz;
int zero_run = 0;
int16_t *zbin_boost_ptr = zbin_boost_orig_ptr;
vpx_memset(qcoeff_ptr, 0, n_coeffs*sizeof(int16_t));
vpx_memset(dqcoeff_ptr, 0, n_coeffs*sizeof(int16_t));
eob = -1;
if (!skip_block) {
for (i = 0; i < n_coeffs; i++) {
rc = scan[i];
z = coeff_ptr[rc] * mul;
zbin = (zbin_ptr[rc != 0] + zbin_boost_ptr[zero_run] + zbin_oq_value);
zero_run += (zero_run < 15);
sz = (z >> 31); // sign of z
x = (z ^ sz) - sz; // x = abs(z)
if (x >= zbin) {
x += (round_ptr[rc != 0]);
y = ((int)(((int)(x * quant_ptr[rc != 0]) >> 16) + x))
>> quant_shift_ptr[rc != 0]; // quantize (x)
x = (y ^ sz) - sz; // get the sign back
qcoeff_ptr[rc] = x; // write to destination
dqcoeff_ptr[rc] = x * dequant_ptr[rc != 0] / mul; // dequantized value
if (y) {
eob = i; // last nonzero coeffs
zero_run = 0;
}
}
}
}
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
*eob_ptr = eob + 1;
}
void vp9_quantize(MACROBLOCK *mb, int plane, int block, int n_coeffs,
TX_TYPE tx_type) {
MACROBLOCKD *const xd = &mb->e_mbd;
const int mul = n_coeffs == 1024 ? 2 : 1;
const int *scan;
// These contexts may be available in the caller
switch (n_coeffs) {
case 4 * 4:
scan = get_scan_4x4(tx_type);
break;
case 8 * 8:
scan = get_scan_8x8(tx_type);
break;
case 16 * 16:
scan = get_scan_16x16(tx_type);
break;
default:
scan = vp9_default_zig_zag1d_32x32;
break;
}
quantize(mb->plane[plane].zrun_zbin_boost,
BLOCK_OFFSET(mb->plane[plane].coeff, block, 16),
n_coeffs, mb->skip_block,
mb->plane[plane].zbin,
mb->plane[plane].round,
mb->plane[plane].quant,
mb->plane[plane].quant_shift,
BLOCK_OFFSET(xd->plane[plane].qcoeff, block, 16),
BLOCK_OFFSET(xd->plane[plane].dqcoeff, block, 16),
xd->plane[plane].dequant,
mb->plane[plane].zbin_extra,
&xd->plane[plane].eobs[block],
scan, mul);
}
void vp9_regular_quantize_b_4x4(MACROBLOCK *mb, int b_idx, TX_TYPE tx_type,
int y_blocks) {
MACROBLOCKD *const xd = &mb->e_mbd;
const struct plane_block_idx pb_idx = plane_block_idx(y_blocks, b_idx);
const int *pt_scan = get_scan_4x4(tx_type);
quantize(mb->plane[pb_idx.plane].zrun_zbin_boost,
BLOCK_OFFSET(mb->plane[pb_idx.plane].coeff, pb_idx.block, 16),
16, mb->skip_block,
mb->plane[pb_idx.plane].zbin,
mb->plane[pb_idx.plane].round,
mb->plane[pb_idx.plane].quant,
mb->plane[pb_idx.plane].quant_shift,
BLOCK_OFFSET(xd->plane[pb_idx.plane].qcoeff, pb_idx.block, 16),
BLOCK_OFFSET(xd->plane[pb_idx.plane].dqcoeff, pb_idx.block, 16),
xd->plane[pb_idx.plane].dequant,
mb->plane[pb_idx.plane].zbin_extra,
&xd->plane[pb_idx.plane].eobs[pb_idx.block],
pt_scan, 1);
}
static void invert_quant(int16_t *quant, uint8_t *shift, int d) {
unsigned t;
int l;
t = d;
for (l = 0; t > 1; l++)
t >>= 1;
t = 1 + (1 << (16 + l)) / d;
*quant = (int16_t)(t - (1 << 16));
*shift = l;
}
void vp9_init_quantizer(VP9_COMP *cpi) {
int i;
int quant_val;
int quant_uv_val;
int q;
static const int zbin_boost[16] = { 0, 0, 0, 8, 8, 8, 10, 12,
14, 16, 20, 24, 28, 32, 36, 40 };
for (q = 0; q < QINDEX_RANGE; q++) {
int qzbin_factor = (vp9_dc_quant(q, 0) < 148) ? 84 : 80;
int qrounding_factor = 48;
if (q == 0) {
qzbin_factor = 64;
qrounding_factor = 64;
}
// dc values
quant_val = vp9_dc_quant(q, cpi->common.y_dc_delta_q);
invert_quant(cpi->y_quant[q] + 0, cpi->y_quant_shift[q] + 0, quant_val);
cpi->y_zbin[q][0] = ROUND_POWER_OF_TWO(qzbin_factor * quant_val, 7);
cpi->y_round[q][0] = (qrounding_factor * quant_val) >> 7;
cpi->common.y_dequant[q][0] = quant_val;
cpi->zrun_zbin_boost_y[q][0] = (quant_val * zbin_boost[0]) >> 7;
quant_val = vp9_dc_quant(q, cpi->common.uv_dc_delta_q);
invert_quant(cpi->uv_quant[q] + 0, cpi->uv_quant_shift[q] + 0, quant_val);
cpi->uv_zbin[q][0] = ROUND_POWER_OF_TWO(qzbin_factor * quant_val, 7);
cpi->uv_round[q][0] = (qrounding_factor * quant_val) >> 7;
cpi->common.uv_dequant[q][0] = quant_val;
cpi->zrun_zbin_boost_uv[q][0] = (quant_val * zbin_boost[0]) >> 7;
quant_val = vp9_ac_quant(q, 0);
cpi->common.y_dequant[q][1] = quant_val;
quant_uv_val = vp9_ac_quant(q, cpi->common.uv_ac_delta_q);
cpi->common.uv_dequant[q][1] = quant_uv_val;
// all the 4x4 ac values =;
for (i = 1; i < 16; i++) {
int rc = vp9_default_zig_zag1d_4x4[i];
invert_quant(cpi->y_quant[q] + rc, cpi->y_quant_shift[q] + rc, quant_val);
cpi->y_zbin[q][rc] = ROUND_POWER_OF_TWO(qzbin_factor * quant_val, 7);
cpi->y_round[q][rc] = (qrounding_factor * quant_val) >> 7;
cpi->zrun_zbin_boost_y[q][i] =
ROUND_POWER_OF_TWO(quant_val * zbin_boost[i], 7);
invert_quant(cpi->uv_quant[q] + rc, cpi->uv_quant_shift[q] + rc,
quant_uv_val);
cpi->uv_zbin[q][rc] = ROUND_POWER_OF_TWO(qzbin_factor * quant_uv_val, 7);
cpi->uv_round[q][rc] = (qrounding_factor * quant_uv_val) >> 7;
cpi->zrun_zbin_boost_uv[q][i] =
ROUND_POWER_OF_TWO(quant_uv_val * zbin_boost[i], 7);
}
}
}
void vp9_mb_init_quantizer(VP9_COMP *cpi, MACROBLOCK *x) {
int i;
MACROBLOCKD *xd = &x->e_mbd;
int zbin_extra;
int segment_id = xd->mode_info_context->mbmi.segment_id;
const int qindex = vp9_get_qindex(xd, segment_id, cpi->common.base_qindex);
// Y
zbin_extra = (cpi->common.y_dequant[qindex][1] *
(cpi->zbin_mode_boost + x->act_zbin_adj)) >> 7;
x->plane[0].quant = cpi->y_quant[qindex];
x->plane[0].quant_shift = cpi->y_quant_shift[qindex];
x->plane[0].zbin = cpi->y_zbin[qindex];
x->plane[0].round = cpi->y_round[qindex];
x->plane[0].zrun_zbin_boost = cpi->zrun_zbin_boost_y[qindex];
x->plane[0].zbin_extra = (int16_t)zbin_extra;
x->e_mbd.plane[0].dequant = cpi->common.y_dequant[qindex];
// UV
zbin_extra = (cpi->common.uv_dequant[qindex][1] *
(cpi->zbin_mode_boost + x->act_zbin_adj)) >> 7;
for (i = 1; i < 3; i++) {
x->plane[i].quant = cpi->uv_quant[qindex];
x->plane[i].quant_shift = cpi->uv_quant_shift[qindex];
x->plane[i].zbin = cpi->uv_zbin[qindex];
x->plane[i].round = cpi->uv_round[qindex];
x->plane[i].zrun_zbin_boost = cpi->zrun_zbin_boost_uv[qindex];
x->plane[i].zbin_extra = (int16_t)zbin_extra;
x->e_mbd.plane[i].dequant = cpi->common.uv_dequant[qindex];
}
x->skip_block = vp9_segfeature_active(xd, segment_id, SEG_LVL_SKIP);
/* save this macroblock QIndex for vp9_update_zbin_extra() */
x->e_mbd.q_index = qindex;
}
void vp9_update_zbin_extra(VP9_COMP *cpi, MACROBLOCK *x) {
const int qindex = x->e_mbd.q_index;
const int y_zbin_extra = (cpi->common.y_dequant[qindex][1] *
(cpi->zbin_mode_boost + x->act_zbin_adj)) >> 7;
const int uv_zbin_extra = (cpi->common.uv_dequant[qindex][1] *
(cpi->zbin_mode_boost + x->act_zbin_adj)) >> 7;
x->plane[0].zbin_extra = (int16_t)y_zbin_extra;
x->plane[1].zbin_extra = (int16_t)uv_zbin_extra;
x->plane[2].zbin_extra = (int16_t)uv_zbin_extra;
}
void vp9_frame_init_quantizer(VP9_COMP *cpi) {
// Clear Zbin mode boost for default case
cpi->zbin_mode_boost = 0;
// MB level quantizer setup
vp9_mb_init_quantizer(cpi, &cpi->mb);
}
void vp9_set_quantizer(struct VP9_COMP *cpi, int Q) {
VP9_COMMON *cm = &cpi->common;
cm->base_qindex = Q;
// if any of the delta_q values are changing update flag will
// have to be set.
cm->y_dc_delta_q = 0;
cm->uv_dc_delta_q = 0;
cm->uv_ac_delta_q = 0;
// quantizer has to be reinitialized if any delta_q changes.
// As there are not any here for now this is inactive code.
// if(update)
// vp9_init_quantizer(cpi);
}