2010-05-18 19:58:33 +04:00
|
|
|
/*
|
2010-09-09 16:16:39 +04:00
|
|
|
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
|
2010-05-18 19:58:33 +04:00
|
|
|
*
|
2010-06-18 20:39:21 +04:00
|
|
|
* Use of this source code is governed by a BSD-style license
|
2010-06-05 00:19:40 +04:00
|
|
|
* that can be found in the LICENSE file in the root of the source
|
|
|
|
* tree. An additional intellectual property rights grant can be found
|
2010-06-18 20:39:21 +04:00
|
|
|
* in the file PATENTS. All contributing project authors may
|
2010-06-05 00:19:40 +04:00
|
|
|
* be found in the AUTHORS file in the root of the source tree.
|
2010-05-18 19:58:33 +04:00
|
|
|
*/
|
|
|
|
|
|
|
|
|
2012-11-28 01:59:17 +04:00
|
|
|
#include "vp9/common/vp9_blockd.h"
|
2012-11-28 22:41:40 +04:00
|
|
|
#include "vp9/decoder/vp9_onyxd_int.h"
|
2010-05-18 19:58:33 +04:00
|
|
|
#include "vpx_mem/vpx_mem.h"
|
|
|
|
#include "vpx_ports/mem.h"
|
2012-11-28 22:41:40 +04:00
|
|
|
#include "vp9/decoder/vp9_detokenize.h"
|
2012-11-28 01:59:17 +04:00
|
|
|
#include "vp9/common/vp9_seg_common.h"
|
2011-10-05 14:26:00 +04:00
|
|
|
|
2010-05-18 19:58:33 +04:00
|
|
|
#define EOB_CONTEXT_NODE 0
|
|
|
|
#define ZERO_CONTEXT_NODE 1
|
|
|
|
#define ONE_CONTEXT_NODE 2
|
|
|
|
#define LOW_VAL_CONTEXT_NODE 3
|
|
|
|
#define TWO_CONTEXT_NODE 4
|
|
|
|
#define THREE_CONTEXT_NODE 5
|
|
|
|
#define HIGH_LOW_CONTEXT_NODE 6
|
|
|
|
#define CAT_ONE_CONTEXT_NODE 7
|
|
|
|
#define CAT_THREEFOUR_CONTEXT_NODE 8
|
|
|
|
#define CAT_THREE_CONTEXT_NODE 9
|
|
|
|
#define CAT_FIVE_CONTEXT_NODE 10
|
|
|
|
|
2011-07-26 01:11:24 +04:00
|
|
|
#define CAT1_MIN_VAL 5
|
|
|
|
#define CAT2_MIN_VAL 7
|
|
|
|
#define CAT3_MIN_VAL 11
|
|
|
|
#define CAT4_MIN_VAL 19
|
|
|
|
#define CAT5_MIN_VAL 35
|
|
|
|
#define CAT6_MIN_VAL 67
|
|
|
|
#define CAT1_PROB0 159
|
|
|
|
#define CAT2_PROB0 145
|
|
|
|
#define CAT2_PROB1 165
|
|
|
|
|
|
|
|
#define CAT3_PROB0 140
|
|
|
|
#define CAT3_PROB1 148
|
|
|
|
#define CAT3_PROB2 173
|
|
|
|
|
|
|
|
#define CAT4_PROB0 135
|
|
|
|
#define CAT4_PROB1 140
|
|
|
|
#define CAT4_PROB2 155
|
|
|
|
#define CAT4_PROB3 176
|
|
|
|
|
|
|
|
#define CAT5_PROB0 130
|
|
|
|
#define CAT5_PROB1 134
|
|
|
|
#define CAT5_PROB2 141
|
|
|
|
#define CAT5_PROB3 157
|
|
|
|
#define CAT5_PROB4 180
|
|
|
|
|
2012-12-19 03:31:19 +04:00
|
|
|
static const vp9_prob cat6_prob[15] = {
|
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
2012-12-08 02:45:05 +04:00
|
|
|
254, 254, 254, 252, 249, 243, 230, 196, 177, 153, 140, 133, 130, 129, 0
|
|
|
|
};
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-12-19 03:31:19 +04:00
|
|
|
DECLARE_ALIGNED(16, extern const uint8_t, vp9_norm[256]);
|
2012-03-19 22:02:04 +04:00
|
|
|
|
2013-03-28 01:03:56 +04:00
|
|
|
static int16_t get_signed(BOOL_DECODER *br, int16_t value_to_sign) {
|
2012-11-16 22:48:23 +04:00
|
|
|
return decode_bool(br, 128) ? -value_to_sign : value_to_sign;
|
2012-07-10 20:36:56 +04:00
|
|
|
}
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2013-03-28 01:03:56 +04:00
|
|
|
|
2012-11-28 03:51:06 +04:00
|
|
|
#define INCREMENT_COUNT(token) \
|
|
|
|
do { \
|
2013-03-25 23:28:24 +04:00
|
|
|
coef_counts[type][ref][get_coef_band(scan, txfm_size, c)] \
|
|
|
|
[pt][token]++; \
|
2013-03-27 03:46:09 +04:00
|
|
|
token_cache[c] = token; \
|
2013-03-25 23:28:24 +04:00
|
|
|
pt = vp9_get_coef_context(scan, nb, pad, token_cache, \
|
2013-03-28 20:24:29 +04:00
|
|
|
c + 1, default_eob); \
|
2012-11-17 22:35:47 +04:00
|
|
|
} while (0)
|
|
|
|
|
2013-02-20 22:16:24 +04:00
|
|
|
#if CONFIG_CODE_NONZEROCOUNT
|
2012-11-17 22:35:47 +04:00
|
|
|
#define WRITE_COEF_CONTINUE(val, token) \
|
2012-08-03 04:03:14 +04:00
|
|
|
{ \
|
2013-03-28 01:03:56 +04:00
|
|
|
qcoeff_ptr[scan[c]] = get_signed(br, val); \
|
2012-11-17 22:35:47 +04:00
|
|
|
INCREMENT_COUNT(token); \
|
2012-07-14 02:21:29 +04:00
|
|
|
c++; \
|
2013-03-28 01:03:56 +04:00
|
|
|
nzc++; \
|
2012-07-14 02:21:29 +04:00
|
|
|
continue; \
|
|
|
|
}
|
2013-02-20 22:16:24 +04:00
|
|
|
#else
|
2013-03-28 01:03:56 +04:00
|
|
|
#define WRITE_COEF_CONTINUE(val, token) \
|
|
|
|
{ \
|
|
|
|
qcoeff_ptr[scan[c]] = get_signed(br, val); \
|
|
|
|
INCREMENT_COUNT(token); \
|
|
|
|
c++; \
|
|
|
|
continue; \
|
2013-02-20 22:16:24 +04:00
|
|
|
}
|
|
|
|
#endif // CONFIG_CODE_NONZEROCOUNT
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-08-03 04:03:14 +04:00
|
|
|
#define ADJUST_COEF(prob, bits_count) \
|
|
|
|
do { \
|
2012-11-01 01:40:53 +04:00
|
|
|
if (vp9_read(br, prob)) \
|
2013-03-28 01:03:56 +04:00
|
|
|
val += 1 << bits_count; \
|
2012-07-14 02:21:29 +04:00
|
|
|
} while (0);
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-10-31 04:53:32 +04:00
|
|
|
static int decode_coefs(VP9D_COMP *dx, const MACROBLOCKD *xd,
|
2013-02-12 05:21:37 +04:00
|
|
|
BOOL_DECODER* const br, int block_idx,
|
2013-03-06 03:18:06 +04:00
|
|
|
PLANE_TYPE type, int seg_eob, int16_t *qcoeff_ptr,
|
|
|
|
TX_SIZE txfm_size) {
|
2013-02-12 05:21:37 +04:00
|
|
|
ENTROPY_CONTEXT* const A0 = (ENTROPY_CONTEXT *) xd->above_context;
|
|
|
|
ENTROPY_CONTEXT* const L0 = (ENTROPY_CONTEXT *) xd->left_context;
|
2013-03-05 02:12:17 +04:00
|
|
|
int aidx, lidx;
|
|
|
|
ENTROPY_CONTEXT above_ec, left_ec;
|
2012-07-14 02:21:29 +04:00
|
|
|
FRAME_CONTEXT *const fc = &dx->common.fc;
|
2013-03-27 03:46:09 +04:00
|
|
|
int pt, c = 0, pad, default_eob;
|
2012-12-08 04:09:59 +04:00
|
|
|
vp9_coeff_probs *coef_probs;
|
|
|
|
vp9_prob *prob;
|
|
|
|
vp9_coeff_count *coef_counts;
|
2013-02-20 01:36:38 +04:00
|
|
|
const int ref = xd->mode_info_context->mbmi.ref_frame != INTRA_FRAME;
|
2013-02-20 22:16:24 +04:00
|
|
|
#if CONFIG_CODE_NONZEROCOUNT
|
2013-03-27 02:23:30 +04:00
|
|
|
const int nzc_used = get_nzc_used(txfm_size);
|
2013-02-20 22:16:24 +04:00
|
|
|
uint16_t nzc = 0;
|
2013-03-27 02:23:30 +04:00
|
|
|
uint16_t nzc_expected =
|
|
|
|
nzc_used ? xd->mode_info_context->mbmi.nzcs[block_idx] : 0;
|
2013-02-20 22:16:24 +04:00
|
|
|
#endif
|
2013-03-27 03:46:09 +04:00
|
|
|
const int *scan, *nb;
|
|
|
|
uint8_t token_cache[1024];
|
2012-08-03 04:03:14 +04:00
|
|
|
|
2013-03-05 02:12:17 +04:00
|
|
|
if (xd->mode_info_context->mbmi.sb_type == BLOCK_SIZE_SB64X64) {
|
|
|
|
aidx = vp9_block2above_sb64[txfm_size][block_idx];
|
|
|
|
lidx = vp9_block2left_sb64[txfm_size][block_idx];
|
|
|
|
} else if (xd->mode_info_context->mbmi.sb_type == BLOCK_SIZE_SB32X32) {
|
|
|
|
aidx = vp9_block2above_sb[txfm_size][block_idx];
|
|
|
|
lidx = vp9_block2left_sb[txfm_size][block_idx];
|
|
|
|
} else {
|
|
|
|
aidx = vp9_block2above[txfm_size][block_idx];
|
|
|
|
lidx = vp9_block2left[txfm_size][block_idx];
|
|
|
|
}
|
|
|
|
|
2012-11-18 06:32:33 +04:00
|
|
|
switch (txfm_size) {
|
2012-08-22 02:08:54 +04:00
|
|
|
default:
|
2013-03-06 03:18:06 +04:00
|
|
|
case TX_4X4: {
|
2013-03-25 23:30:00 +04:00
|
|
|
const TX_TYPE tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
|
|
|
|
get_tx_type_4x4(xd, block_idx) : DCT_DCT;
|
2013-03-06 03:18:06 +04:00
|
|
|
switch (tx_type) {
|
|
|
|
default:
|
|
|
|
scan = vp9_default_zig_zag1d_4x4;
|
|
|
|
break;
|
|
|
|
case ADST_DCT:
|
|
|
|
scan = vp9_row_scan_4x4;
|
|
|
|
break;
|
|
|
|
case DCT_ADST:
|
|
|
|
scan = vp9_col_scan_4x4;
|
|
|
|
break;
|
|
|
|
}
|
2013-03-05 02:12:17 +04:00
|
|
|
above_ec = A0[aidx] != 0;
|
|
|
|
left_ec = L0[lidx] != 0;
|
2013-02-20 01:36:38 +04:00
|
|
|
coef_probs = fc->coef_probs_4x4;
|
|
|
|
coef_counts = fc->coef_counts_4x4;
|
2013-03-27 03:46:09 +04:00
|
|
|
default_eob = 16;
|
2012-08-03 04:03:14 +04:00
|
|
|
break;
|
2013-03-06 03:18:06 +04:00
|
|
|
}
|
2013-03-25 23:30:00 +04:00
|
|
|
case TX_8X8: {
|
|
|
|
const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
|
|
|
|
const int sz = 3 + sb_type, x = block_idx & ((1 << sz) - 1);
|
|
|
|
const int y = block_idx - x;
|
|
|
|
const TX_TYPE tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
|
|
|
|
get_tx_type_8x8(xd, y + (x >> 1)) : DCT_DCT;
|
|
|
|
switch (tx_type) {
|
|
|
|
default:
|
|
|
|
scan = vp9_default_zig_zag1d_8x8;
|
|
|
|
break;
|
|
|
|
case ADST_DCT:
|
|
|
|
scan = vp9_row_scan_8x8;
|
|
|
|
break;
|
|
|
|
case DCT_ADST:
|
|
|
|
scan = vp9_col_scan_8x8;
|
|
|
|
break;
|
|
|
|
}
|
2013-02-20 01:36:38 +04:00
|
|
|
coef_probs = fc->coef_probs_8x8;
|
|
|
|
coef_counts = fc->coef_counts_8x8;
|
2013-02-15 22:15:42 +04:00
|
|
|
above_ec = (A0[aidx] + A0[aidx + 1]) != 0;
|
|
|
|
left_ec = (L0[lidx] + L0[lidx + 1]) != 0;
|
2013-03-27 03:46:09 +04:00
|
|
|
default_eob = 64;
|
2012-08-03 04:03:14 +04:00
|
|
|
break;
|
2013-03-25 23:30:00 +04:00
|
|
|
}
|
|
|
|
case TX_16X16: {
|
|
|
|
const BLOCK_SIZE_TYPE sb_type = xd->mode_info_context->mbmi.sb_type;
|
|
|
|
const int sz = 4 + sb_type, x = block_idx & ((1 << sz) - 1);
|
|
|
|
const int y = block_idx - x;
|
|
|
|
const TX_TYPE tx_type = (type == PLANE_TYPE_Y_WITH_DC) ?
|
|
|
|
get_tx_type_16x16(xd, y + (x >> 2)) : DCT_DCT;
|
|
|
|
switch (tx_type) {
|
|
|
|
default:
|
|
|
|
scan = vp9_default_zig_zag1d_16x16;
|
|
|
|
break;
|
|
|
|
case ADST_DCT:
|
|
|
|
scan = vp9_row_scan_16x16;
|
|
|
|
break;
|
|
|
|
case DCT_ADST:
|
|
|
|
scan = vp9_col_scan_16x16;
|
|
|
|
break;
|
|
|
|
}
|
2013-02-20 01:36:38 +04:00
|
|
|
coef_probs = fc->coef_probs_16x16;
|
|
|
|
coef_counts = fc->coef_counts_16x16;
|
2013-02-12 05:21:37 +04:00
|
|
|
if (type == PLANE_TYPE_UV) {
|
|
|
|
ENTROPY_CONTEXT *A1 = (ENTROPY_CONTEXT *) (xd->above_context + 1);
|
|
|
|
ENTROPY_CONTEXT *L1 = (ENTROPY_CONTEXT *) (xd->left_context + 1);
|
|
|
|
above_ec = (A0[aidx] + A0[aidx + 1] + A1[aidx] + A1[aidx + 1]) != 0;
|
|
|
|
left_ec = (L0[lidx] + L0[lidx + 1] + L1[lidx] + L1[lidx + 1]) != 0;
|
2013-02-15 22:15:42 +04:00
|
|
|
} else {
|
2013-02-12 05:21:37 +04:00
|
|
|
above_ec = (A0[aidx] + A0[aidx + 1] + A0[aidx + 2] + A0[aidx + 3]) != 0;
|
|
|
|
left_ec = (L0[lidx] + L0[lidx + 1] + L0[lidx + 2] + L0[lidx + 3]) != 0;
|
|
|
|
}
|
2013-03-27 03:46:09 +04:00
|
|
|
default_eob = 256;
|
2012-08-03 04:03:14 +04:00
|
|
|
break;
|
2013-03-25 23:30:00 +04:00
|
|
|
}
|
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
2012-12-08 02:45:05 +04:00
|
|
|
case TX_32X32:
|
2013-03-06 03:18:06 +04:00
|
|
|
scan = vp9_default_zig_zag1d_32x32;
|
2012-12-08 04:09:59 +04:00
|
|
|
coef_probs = fc->coef_probs_32x32;
|
|
|
|
coef_counts = fc->coef_counts_32x32;
|
2013-02-12 05:21:37 +04:00
|
|
|
if (type == PLANE_TYPE_UV) {
|
|
|
|
ENTROPY_CONTEXT *A1 = (ENTROPY_CONTEXT *) (xd->above_context + 1);
|
|
|
|
ENTROPY_CONTEXT *L1 = (ENTROPY_CONTEXT *) (xd->left_context + 1);
|
|
|
|
ENTROPY_CONTEXT *A2 = (ENTROPY_CONTEXT *) (xd->above_context + 2);
|
|
|
|
ENTROPY_CONTEXT *L2 = (ENTROPY_CONTEXT *) (xd->left_context + 2);
|
|
|
|
ENTROPY_CONTEXT *A3 = (ENTROPY_CONTEXT *) (xd->above_context + 3);
|
|
|
|
ENTROPY_CONTEXT *L3 = (ENTROPY_CONTEXT *) (xd->left_context + 3);
|
|
|
|
above_ec = (A0[aidx] + A0[aidx + 1] + A1[aidx] + A1[aidx + 1] +
|
|
|
|
A2[aidx] + A2[aidx + 1] + A3[aidx] + A3[aidx + 1]) != 0;
|
|
|
|
left_ec = (L0[lidx] + L0[lidx + 1] + L1[lidx] + L1[lidx + 1] +
|
|
|
|
L2[lidx] + L2[lidx + 1] + L3[lidx] + L3[lidx + 1]) != 0;
|
2013-02-15 22:15:42 +04:00
|
|
|
} else {
|
2013-02-12 05:21:37 +04:00
|
|
|
ENTROPY_CONTEXT *A1 = (ENTROPY_CONTEXT *) (xd->above_context + 1);
|
|
|
|
ENTROPY_CONTEXT *L1 = (ENTROPY_CONTEXT *) (xd->left_context + 1);
|
|
|
|
above_ec = (A0[aidx] + A0[aidx + 1] + A0[aidx + 2] + A0[aidx + 3] +
|
|
|
|
A1[aidx] + A1[aidx + 1] + A1[aidx + 2] + A1[aidx + 3]) != 0;
|
|
|
|
left_ec = (L0[lidx] + L0[lidx + 1] + L0[lidx + 2] + L0[lidx + 3] +
|
|
|
|
L1[lidx] + L1[lidx + 1] + L1[lidx + 2] + L1[lidx + 3]) != 0;
|
|
|
|
}
|
2013-03-27 03:46:09 +04:00
|
|
|
default_eob = 1024;
|
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
2012-12-08 02:45:05 +04:00
|
|
|
break;
|
2012-08-03 04:03:14 +04:00
|
|
|
}
|
2012-07-14 02:21:29 +04:00
|
|
|
|
2013-02-12 05:21:37 +04:00
|
|
|
VP9_COMBINEENTROPYCONTEXTS(pt, above_ec, left_ec);
|
2013-03-27 03:46:09 +04:00
|
|
|
nb = vp9_get_coef_neighbors_handle(scan, &pad);
|
|
|
|
|
2012-07-14 02:21:29 +04:00
|
|
|
while (1) {
|
|
|
|
int val;
|
|
|
|
const uint8_t *cat6 = cat6_prob;
|
2013-02-23 05:27:34 +04:00
|
|
|
|
|
|
|
if (c >= seg_eob)
|
|
|
|
break;
|
2013-02-20 22:16:24 +04:00
|
|
|
#if CONFIG_CODE_NONZEROCOUNT
|
2013-03-27 02:23:30 +04:00
|
|
|
if (nzc_used && nzc == nzc_expected)
|
2013-02-20 22:16:24 +04:00
|
|
|
break;
|
|
|
|
#endif
|
2013-03-25 23:28:24 +04:00
|
|
|
prob = coef_probs[type][ref][get_coef_band(scan, txfm_size, c)][pt];
|
2013-03-27 03:46:09 +04:00
|
|
|
fc->eob_branch_counts[txfm_size][type][ref]
|
2013-03-25 23:28:24 +04:00
|
|
|
[get_coef_band(scan, txfm_size, c)][pt]++;
|
2013-03-27 02:23:30 +04:00
|
|
|
#if CONFIG_CODE_NONZEROCOUNT
|
|
|
|
if (!nzc_used)
|
2013-02-20 22:16:24 +04:00
|
|
|
#endif
|
2013-03-27 02:23:30 +04:00
|
|
|
if (!vp9_read(br, prob[EOB_CONTEXT_NODE]))
|
|
|
|
break;
|
2012-08-03 04:03:14 +04:00
|
|
|
SKIP_START:
|
2013-02-23 05:27:34 +04:00
|
|
|
if (c >= seg_eob)
|
|
|
|
break;
|
2013-02-20 22:16:24 +04:00
|
|
|
#if CONFIG_CODE_NONZEROCOUNT
|
2013-03-27 02:23:30 +04:00
|
|
|
if (nzc_used && nzc == nzc_expected)
|
2013-02-20 22:16:24 +04:00
|
|
|
break;
|
|
|
|
// decode zero node only if there are zeros left
|
2013-03-27 02:23:30 +04:00
|
|
|
if (!nzc_used || seg_eob - nzc_expected - c + nzc > 0)
|
2013-02-20 22:16:24 +04:00
|
|
|
#endif
|
2012-11-01 01:40:53 +04:00
|
|
|
if (!vp9_read(br, prob[ZERO_CONTEXT_NODE])) {
|
2012-11-17 22:35:47 +04:00
|
|
|
INCREMENT_COUNT(ZERO_TOKEN);
|
2012-07-14 02:21:29 +04:00
|
|
|
++c;
|
2013-03-25 23:28:24 +04:00
|
|
|
prob = coef_probs[type][ref][get_coef_band(scan, txfm_size, c)][pt];
|
2012-07-14 02:21:29 +04:00
|
|
|
goto SKIP_START;
|
|
|
|
}
|
|
|
|
// ONE_CONTEXT_NODE_0_
|
2012-11-01 01:40:53 +04:00
|
|
|
if (!vp9_read(br, prob[ONE_CONTEXT_NODE])) {
|
2012-11-17 23:07:23 +04:00
|
|
|
WRITE_COEF_CONTINUE(1, ONE_TOKEN);
|
2012-07-14 02:21:29 +04:00
|
|
|
}
|
|
|
|
// LOW_VAL_CONTEXT_NODE_0_
|
2012-11-01 01:40:53 +04:00
|
|
|
if (!vp9_read(br, prob[LOW_VAL_CONTEXT_NODE])) {
|
|
|
|
if (!vp9_read(br, prob[TWO_CONTEXT_NODE])) {
|
2012-11-17 22:35:47 +04:00
|
|
|
WRITE_COEF_CONTINUE(2, TWO_TOKEN);
|
2012-07-14 02:21:29 +04:00
|
|
|
}
|
2012-11-01 01:40:53 +04:00
|
|
|
if (!vp9_read(br, prob[THREE_CONTEXT_NODE])) {
|
2012-11-17 22:35:47 +04:00
|
|
|
WRITE_COEF_CONTINUE(3, THREE_TOKEN);
|
2012-07-14 02:21:29 +04:00
|
|
|
}
|
2012-11-17 22:35:47 +04:00
|
|
|
WRITE_COEF_CONTINUE(4, FOUR_TOKEN);
|
2012-07-14 02:21:29 +04:00
|
|
|
}
|
|
|
|
// HIGH_LOW_CONTEXT_NODE_0_
|
2012-11-01 01:40:53 +04:00
|
|
|
if (!vp9_read(br, prob[HIGH_LOW_CONTEXT_NODE])) {
|
|
|
|
if (!vp9_read(br, prob[CAT_ONE_CONTEXT_NODE])) {
|
2012-07-14 02:21:29 +04:00
|
|
|
val = CAT1_MIN_VAL;
|
|
|
|
ADJUST_COEF(CAT1_PROB0, 0);
|
2012-11-17 22:35:47 +04:00
|
|
|
WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY1);
|
2012-07-14 02:21:29 +04:00
|
|
|
}
|
|
|
|
val = CAT2_MIN_VAL;
|
|
|
|
ADJUST_COEF(CAT2_PROB1, 1);
|
|
|
|
ADJUST_COEF(CAT2_PROB0, 0);
|
2012-11-17 22:35:47 +04:00
|
|
|
WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY2);
|
2012-07-10 20:36:56 +04:00
|
|
|
}
|
2012-07-14 02:21:29 +04:00
|
|
|
// CAT_THREEFOUR_CONTEXT_NODE_0_
|
2012-11-01 01:40:53 +04:00
|
|
|
if (!vp9_read(br, prob[CAT_THREEFOUR_CONTEXT_NODE])) {
|
|
|
|
if (!vp9_read(br, prob[CAT_THREE_CONTEXT_NODE])) {
|
2012-07-14 02:21:29 +04:00
|
|
|
val = CAT3_MIN_VAL;
|
|
|
|
ADJUST_COEF(CAT3_PROB2, 2);
|
|
|
|
ADJUST_COEF(CAT3_PROB1, 1);
|
|
|
|
ADJUST_COEF(CAT3_PROB0, 0);
|
2012-11-17 22:35:47 +04:00
|
|
|
WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY3);
|
2012-07-14 02:21:29 +04:00
|
|
|
}
|
|
|
|
val = CAT4_MIN_VAL;
|
|
|
|
ADJUST_COEF(CAT4_PROB3, 3);
|
|
|
|
ADJUST_COEF(CAT4_PROB2, 2);
|
|
|
|
ADJUST_COEF(CAT4_PROB1, 1);
|
|
|
|
ADJUST_COEF(CAT4_PROB0, 0);
|
2012-11-17 22:35:47 +04:00
|
|
|
WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY4);
|
2012-07-14 02:21:29 +04:00
|
|
|
}
|
|
|
|
// CAT_FIVE_CONTEXT_NODE_0_:
|
2012-11-01 01:40:53 +04:00
|
|
|
if (!vp9_read(br, prob[CAT_FIVE_CONTEXT_NODE])) {
|
2012-07-14 02:21:29 +04:00
|
|
|
val = CAT5_MIN_VAL;
|
|
|
|
ADJUST_COEF(CAT5_PROB4, 4);
|
|
|
|
ADJUST_COEF(CAT5_PROB3, 3);
|
|
|
|
ADJUST_COEF(CAT5_PROB2, 2);
|
|
|
|
ADJUST_COEF(CAT5_PROB1, 1);
|
|
|
|
ADJUST_COEF(CAT5_PROB0, 0);
|
2012-11-17 22:35:47 +04:00
|
|
|
WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY5);
|
2012-07-14 02:21:29 +04:00
|
|
|
}
|
|
|
|
val = 0;
|
|
|
|
while (*cat6) {
|
2012-11-01 01:40:53 +04:00
|
|
|
val = (val << 1) | vp9_read(br, *cat6++);
|
2012-07-14 02:21:29 +04:00
|
|
|
}
|
|
|
|
val += CAT6_MIN_VAL;
|
2012-11-17 22:35:47 +04:00
|
|
|
WRITE_COEF_CONTINUE(val, DCT_VAL_CATEGORY6);
|
2012-07-14 02:21:29 +04:00
|
|
|
}
|
2012-06-25 23:26:09 +04:00
|
|
|
|
2013-03-27 02:23:30 +04:00
|
|
|
#if CONFIG_CODE_NONZEROCOUNT
|
|
|
|
if (!nzc_used)
|
|
|
|
#endif
|
|
|
|
if (c < seg_eob)
|
|
|
|
coef_counts[type][ref][get_coef_band(scan, txfm_size, c)]
|
|
|
|
[pt][DCT_EOB_TOKEN]++;
|
|
|
|
#if CONFIG_CODE_NONZEROCOUNT
|
|
|
|
if (!nzc_used)
|
|
|
|
xd->mode_info_context->mbmi.nzcs[block_idx] = nzc;
|
|
|
|
else
|
|
|
|
assert(nzc == nzc_expected);
|
2013-02-20 22:16:24 +04:00
|
|
|
#endif
|
2012-11-17 22:18:41 +04:00
|
|
|
|
2013-02-23 05:27:34 +04:00
|
|
|
A0[aidx] = L0[lidx] = c > 0;
|
2013-02-15 22:15:42 +04:00
|
|
|
if (txfm_size >= TX_8X8) {
|
2013-02-12 05:21:37 +04:00
|
|
|
A0[aidx + 1] = L0[lidx + 1] = A0[aidx];
|
|
|
|
if (txfm_size >= TX_16X16) {
|
|
|
|
if (type == PLANE_TYPE_UV) {
|
|
|
|
ENTROPY_CONTEXT *A1 = (ENTROPY_CONTEXT *) (xd->above_context + 1);
|
|
|
|
ENTROPY_CONTEXT *L1 = (ENTROPY_CONTEXT *) (xd->left_context + 1);
|
2013-03-05 02:12:17 +04:00
|
|
|
A1[aidx] = A1[aidx + 1] = L1[lidx] = L1[lidx + 1] = A0[aidx];
|
2013-02-12 05:21:37 +04:00
|
|
|
if (txfm_size >= TX_32X32) {
|
|
|
|
ENTROPY_CONTEXT *A2 = (ENTROPY_CONTEXT *) (xd->above_context + 2);
|
|
|
|
ENTROPY_CONTEXT *L2 = (ENTROPY_CONTEXT *) (xd->left_context + 2);
|
|
|
|
ENTROPY_CONTEXT *A3 = (ENTROPY_CONTEXT *) (xd->above_context + 3);
|
|
|
|
ENTROPY_CONTEXT *L3 = (ENTROPY_CONTEXT *) (xd->left_context + 3);
|
|
|
|
A2[aidx] = A2[aidx + 1] = A3[aidx] = A3[aidx + 1] = A0[aidx];
|
|
|
|
L2[lidx] = L2[lidx + 1] = L3[lidx] = L3[lidx + 1] = A0[aidx];
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
A0[aidx + 2] = A0[aidx + 3] = L0[lidx + 2] = L0[lidx + 3] = A0[aidx];
|
|
|
|
if (txfm_size >= TX_32X32) {
|
|
|
|
ENTROPY_CONTEXT *A1 = (ENTROPY_CONTEXT *) (xd->above_context + 1);
|
|
|
|
ENTROPY_CONTEXT *L1 = (ENTROPY_CONTEXT *) (xd->left_context + 1);
|
|
|
|
A1[aidx] = A1[aidx + 1] = A1[aidx + 2] = A1[aidx + 3] = A0[aidx];
|
|
|
|
L1[lidx] = L1[lidx + 1] = L1[lidx + 2] = L1[lidx + 3] = A0[aidx];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2012-07-14 02:21:29 +04:00
|
|
|
return c;
|
2012-07-10 20:36:56 +04:00
|
|
|
}
|
2010-05-18 19:58:33 +04:00
|
|
|
|
2012-11-23 20:17:06 +04:00
|
|
|
static int get_eob(MACROBLOCKD* const xd, int segment_id, int eob_max) {
|
2013-02-23 05:27:34 +04:00
|
|
|
return vp9_get_segdata(xd, segment_id, SEG_LVL_SKIP) ? 0 : eob_max;
|
2012-11-14 21:51:23 +04:00
|
|
|
}
|
|
|
|
|
2013-04-06 02:54:59 +04:00
|
|
|
/* TODO(jkoleszar): Probably best to remove instances that require this,
|
|
|
|
* as the data likely becomes per-plane and stored in the per-plane structures.
|
|
|
|
* This is a stub to work with the existing code.
|
|
|
|
*/
|
|
|
|
static INLINE int block_idx_4x4(MACROBLOCKD* const xd, int block_size_b,
|
|
|
|
int plane, int i) {
|
|
|
|
const int luma_blocks = 1 << block_size_b;
|
|
|
|
assert(xd->plane[0].subsampling_x == 0);
|
|
|
|
assert(xd->plane[0].subsampling_y == 0);
|
|
|
|
assert(xd->plane[1].subsampling_x == 1);
|
|
|
|
assert(xd->plane[1].subsampling_y == 1);
|
|
|
|
assert(xd->plane[2].subsampling_x == 1);
|
|
|
|
assert(xd->plane[2].subsampling_y == 1);
|
|
|
|
return plane == 0 ? i :
|
|
|
|
plane == 1 ? luma_blocks + i :
|
|
|
|
luma_blocks * 5 / 4 + i;
|
|
|
|
}
|
|
|
|
|
|
|
|
static INLINE int decode_block_plane(VP9D_COMP* const pbi,
|
|
|
|
MACROBLOCKD* const xd,
|
|
|
|
BOOL_DECODER* const bc,
|
|
|
|
BLOCK_SIZE_LG2 block_size,
|
|
|
|
int segment_id,
|
|
|
|
int plane,
|
|
|
|
int is_split) {
|
|
|
|
// block and transform sizes, in number of 4x4 blocks log 2 ("*_b")
|
|
|
|
// 4x4=0, 8x8=2, 16x16=4, 32x32=6, 64x64=8
|
|
|
|
const TX_SIZE tx_size = xd->mode_info_context->mbmi.txfm_size;
|
|
|
|
const BLOCK_SIZE_LG2 block_size_b = block_size;
|
|
|
|
const BLOCK_SIZE_LG2 txfrm_size_b = tx_size * 2;
|
|
|
|
|
|
|
|
// subsampled size of the block
|
|
|
|
const int ss_sum = xd->plane[plane].subsampling_x +
|
|
|
|
xd->plane[plane].subsampling_y;
|
|
|
|
const BLOCK_SIZE_LG2 ss_block_size = block_size_b - ss_sum;
|
|
|
|
|
|
|
|
// size of the transform to use. scale the transform down if it's larger
|
|
|
|
// than the size of the subsampled data, or forced externally by the mb mode.
|
|
|
|
const int ss_max = MAX(xd->plane[plane].subsampling_x,
|
|
|
|
xd->plane[plane].subsampling_y);
|
|
|
|
const BLOCK_SIZE_LG2 ss_txfrm_size = txfrm_size_b > ss_block_size || is_split
|
|
|
|
? txfrm_size_b - ss_max * 2
|
|
|
|
: txfrm_size_b;
|
|
|
|
const TX_SIZE ss_tx_size = ss_txfrm_size / 2;
|
|
|
|
|
|
|
|
// TODO(jkoleszar): 1 may not be correct here with larger chroma planes.
|
|
|
|
const int inc = is_split ? 1 : (1 << ss_txfrm_size);
|
|
|
|
|
|
|
|
// find the maximum eob for this transform size, adjusted by segment
|
|
|
|
const int seg_eob = get_eob(xd, segment_id, 16 << ss_txfrm_size);
|
|
|
|
|
2013-03-28 01:03:56 +04:00
|
|
|
int i, eobtotal = 0;
|
|
|
|
|
2013-04-06 02:54:59 +04:00
|
|
|
assert(txfrm_size_b <= block_size_b);
|
|
|
|
assert(ss_txfrm_size <= ss_block_size);
|
2013-04-03 01:50:40 +04:00
|
|
|
|
2013-04-06 02:54:59 +04:00
|
|
|
// step through the block by the size of the transform in use.
|
|
|
|
for (i = 0; i < (1 << ss_block_size); i += inc) {
|
|
|
|
const int block_idx = block_idx_4x4(xd, block_size_b, plane, i);
|
2013-03-28 01:03:56 +04:00
|
|
|
|
2013-04-06 02:54:59 +04:00
|
|
|
const int c = decode_coefs(pbi, xd, bc, block_idx,
|
|
|
|
xd->plane[plane].plane_type, seg_eob,
|
|
|
|
BLOCK_OFFSET(xd->plane[plane].qcoeff, i, 16),
|
|
|
|
ss_tx_size);
|
|
|
|
xd->plane[plane].eobs[i] = c;
|
2013-04-03 01:50:40 +04:00
|
|
|
eobtotal += c;
|
|
|
|
}
|
2013-03-28 01:03:56 +04:00
|
|
|
return eobtotal;
|
|
|
|
}
|
|
|
|
|
2013-04-06 02:54:59 +04:00
|
|
|
static INLINE int decode_blocks_helper(VP9D_COMP* const pbi,
|
|
|
|
MACROBLOCKD* const xd,
|
|
|
|
BOOL_DECODER* const bc,
|
|
|
|
int block_size,
|
|
|
|
int is_split_chroma) {
|
|
|
|
const int segment_id = xd->mode_info_context->mbmi.segment_id;
|
|
|
|
int plane, eobtotal = 0;
|
|
|
|
|
|
|
|
for (plane = 0; plane < MAX_MB_PLANE; plane++) {
|
|
|
|
const int is_split = is_split_chroma &&
|
|
|
|
xd->plane[plane].plane_type == PLANE_TYPE_UV;
|
|
|
|
eobtotal += decode_block_plane(pbi, xd, bc, block_size, segment_id,
|
|
|
|
plane, is_split);
|
2013-03-05 02:12:17 +04:00
|
|
|
}
|
2013-04-06 02:54:59 +04:00
|
|
|
return eobtotal;
|
|
|
|
}
|
|
|
|
|
|
|
|
static INLINE int decode_blocks(VP9D_COMP* const pbi,
|
|
|
|
MACROBLOCKD* const xd,
|
|
|
|
BOOL_DECODER* const bc,
|
|
|
|
int block_size) {
|
|
|
|
const MB_PREDICTION_MODE mode = xd->mode_info_context->mbmi.mode;
|
|
|
|
const TX_SIZE tx_size = xd->mode_info_context->mbmi.txfm_size;
|
|
|
|
return decode_blocks_helper(pbi, xd, bc, block_size,
|
|
|
|
tx_size == TX_8X8 && (mode == I8X8_PRED || mode == SPLITMV));
|
2013-03-05 02:12:17 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
int vp9_decode_sb64_tokens(VP9D_COMP* const pbi,
|
|
|
|
MACROBLOCKD* const xd,
|
|
|
|
BOOL_DECODER* const bc) {
|
2013-04-06 02:54:59 +04:00
|
|
|
return decode_blocks(pbi, xd, bc, BLOCK_64X64_LG2);
|
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
2012-12-08 02:45:05 +04:00
|
|
|
}
|
2012-11-14 21:51:23 +04:00
|
|
|
|
2013-04-06 02:54:59 +04:00
|
|
|
int vp9_decode_sb_tokens(VP9D_COMP* const pbi,
|
|
|
|
MACROBLOCKD* const xd,
|
|
|
|
BOOL_DECODER* const bc) {
|
|
|
|
return decode_blocks(pbi, xd, bc, BLOCK_32X32_LG2);
|
2012-08-03 04:03:14 +04:00
|
|
|
}
|
|
|
|
|
2013-04-06 02:54:59 +04:00
|
|
|
int vp9_decode_mb_tokens(VP9D_COMP* const pbi,
|
|
|
|
MACROBLOCKD* const xd,
|
|
|
|
BOOL_DECODER* const bc) {
|
|
|
|
return decode_blocks(pbi, xd, bc, BLOCK_16X16_LG2);
|
2012-07-11 04:23:38 +04:00
|
|
|
}
|
|
|
|
|
2013-04-06 02:54:59 +04:00
|
|
|
#if CONFIG_NEWBINTRAMODES
|
2012-12-08 04:26:25 +04:00
|
|
|
static int decode_coefs_4x4(VP9D_COMP *dx, MACROBLOCKD *xd,
|
|
|
|
BOOL_DECODER* const bc,
|
2013-03-06 03:18:06 +04:00
|
|
|
PLANE_TYPE type, int i, int seg_eob) {
|
2013-04-04 23:03:27 +04:00
|
|
|
const struct plane_block_idx pb_idx = plane_block_idx(16, i);
|
2013-03-28 01:03:56 +04:00
|
|
|
const int c = decode_coefs(dx, xd, bc, i, type, seg_eob,
|
2013-04-04 23:03:27 +04:00
|
|
|
BLOCK_OFFSET(xd->plane[pb_idx.plane].qcoeff, pb_idx.block, 16), TX_4X4);
|
|
|
|
xd->plane[pb_idx.plane].eobs[pb_idx.block] = c;
|
2012-12-08 04:26:25 +04:00
|
|
|
return c;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int decode_mb_tokens_4x4_uv(VP9D_COMP* const dx,
|
|
|
|
MACROBLOCKD* const xd,
|
|
|
|
BOOL_DECODER* const bc,
|
|
|
|
int seg_eob) {
|
2013-03-28 01:03:56 +04:00
|
|
|
int i, eobtotal = 0;
|
2012-11-22 00:39:55 +04:00
|
|
|
|
2012-12-08 04:26:25 +04:00
|
|
|
// chroma blocks
|
2013-03-28 01:03:56 +04:00
|
|
|
for (i = 16; i < 24; i++)
|
2013-03-06 03:18:06 +04:00
|
|
|
eobtotal += decode_coefs_4x4(dx, xd, bc, PLANE_TYPE_UV, i, seg_eob);
|
2012-11-22 00:39:55 +04:00
|
|
|
|
|
|
|
return eobtotal;
|
|
|
|
}
|
|
|
|
|
2012-12-08 04:26:25 +04:00
|
|
|
int vp9_decode_mb_tokens_4x4_uv(VP9D_COMP* const dx,
|
|
|
|
MACROBLOCKD* const xd,
|
|
|
|
BOOL_DECODER* const bc) {
|
|
|
|
const int segment_id = xd->mode_info_context->mbmi.segment_id;
|
|
|
|
const int seg_eob = get_eob(xd, segment_id, 16);
|
|
|
|
|
|
|
|
return decode_mb_tokens_4x4_uv(dx, xd, bc, seg_eob);
|
|
|
|
}
|
|
|
|
|
2013-04-04 23:32:34 +04:00
|
|
|
int vp9_decode_coefs_4x4(VP9D_COMP *dx, MACROBLOCKD *xd,
|
|
|
|
BOOL_DECODER* const bc,
|
|
|
|
PLANE_TYPE type, int i) {
|
|
|
|
const int segment_id = xd->mode_info_context->mbmi.segment_id;
|
|
|
|
const int seg_eob = get_eob(xd, segment_id, 16);
|
|
|
|
return decode_coefs_4x4(dx, xd, bc, type, i, seg_eob);
|
|
|
|
}
|
|
|
|
#endif
|