vp9_idct.h 6.19 KB
Newer Older
1 2 3 4 5 6 7 8 9 10
/*
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
 *
 *  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.
 */

Dmitry Kovalev's avatar
Dmitry Kovalev committed
11 12
#ifndef VP9_COMMON_VP9_IDCT_H_
#define VP9_COMMON_VP9_IDCT_H_
13

Christian Duvivier's avatar
Christian Duvivier committed
14 15
#include <assert.h>

16
#include "./vpx_config.h"
17
#include "vp9/common/vp9_common.h"
18
#include "vp9/common/vp9_enums.h"
19

20 21 22 23
#ifdef __cplusplus
extern "C" {
#endif

24 25 26
// Constants and Macros used by all idct/dct functions
#define DCT_CONST_BITS 14
#define DCT_CONST_ROUNDING  (1 << (DCT_CONST_BITS - 1))
27

Yaowu Xu's avatar
Yaowu Xu committed
28 29
#define UNIT_QUANT_SHIFT 2
#define UNIT_QUANT_FACTOR (1 << UNIT_QUANT_SHIFT)
30

Yunqing Wang's avatar
Yunqing Wang committed
31
#define pair_set_epi16(a, b) \
Yaowu Xu's avatar
Yaowu Xu committed
32
  _mm_set_epi16(b, a, b, a, b, a, b, a)
Yunqing Wang's avatar
Yunqing Wang committed
33

34 35 36
#define dual_set_epi16(a, b) \
  _mm_set_epi16(b, b, b, b, a, a, a, a)

37 38 39 40
// Constants:
//  for (int i = 1; i< 32; ++i)
//    printf("static const int cospi_%d_64 = %.0f;\n", i,
//           round(16384 * cos(i*M_PI/64)));
Dmitry Kovalev's avatar
Dmitry Kovalev committed
41
// Note: sin(k*Pi/64) = cos((32-k)*Pi/64)
42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72
static const tran_high_t cospi_1_64  = 16364;
static const tran_high_t cospi_2_64  = 16305;
static const tran_high_t cospi_3_64  = 16207;
static const tran_high_t cospi_4_64  = 16069;
static const tran_high_t cospi_5_64  = 15893;
static const tran_high_t cospi_6_64  = 15679;
static const tran_high_t cospi_7_64  = 15426;
static const tran_high_t cospi_8_64  = 15137;
static const tran_high_t cospi_9_64  = 14811;
static const tran_high_t cospi_10_64 = 14449;
static const tran_high_t cospi_11_64 = 14053;
static const tran_high_t cospi_12_64 = 13623;
static const tran_high_t cospi_13_64 = 13160;
static const tran_high_t cospi_14_64 = 12665;
static const tran_high_t cospi_15_64 = 12140;
static const tran_high_t cospi_16_64 = 11585;
static const tran_high_t cospi_17_64 = 11003;
static const tran_high_t cospi_18_64 = 10394;
static const tran_high_t cospi_19_64 = 9760;
static const tran_high_t cospi_20_64 = 9102;
static const tran_high_t cospi_21_64 = 8423;
static const tran_high_t cospi_22_64 = 7723;
static const tran_high_t cospi_23_64 = 7005;
static const tran_high_t cospi_24_64 = 6270;
static const tran_high_t cospi_25_64 = 5520;
static const tran_high_t cospi_26_64 = 4756;
static const tran_high_t cospi_27_64 = 3981;
static const tran_high_t cospi_28_64 = 3196;
static const tran_high_t cospi_29_64 = 2404;
static const tran_high_t cospi_30_64 = 1606;
static const tran_high_t cospi_31_64 = 804;
73

Jingning Han's avatar
Jingning Han committed
74
//  16384 * sqrt(2) * sin(kPi/9) * 2 / 3
75 76 77 78 79
static const tran_high_t sinpi_1_9 = 5283;
static const tran_high_t sinpi_2_9 = 9929;
static const tran_high_t sinpi_3_9 = 13377;
static const tran_high_t sinpi_4_9 = 15212;

80
static INLINE tran_low_t check_range(tran_high_t input) {
81 82 83 84 85 86 87
#if CONFIG_VP9_HIGHBITDEPTH
  // For valid highbitdepth VP9 streams, intermediate stage coefficients will
  // stay within the ranges:
  // - 8 bit: signed 16 bit integer
  // - 10 bit: signed 18 bit integer
  // - 12 bit: signed 20 bit integer
#elif CONFIG_COEFFICIENT_RANGE_CHECKING
88 89 90 91 92 93
  // For valid VP9 input streams, intermediate stage coefficients should always
  // stay within the range of a signed 16 bit integer. Coefficients can go out
  // of this range for invalid/corrupt VP9 streams. However, strictly checking
  // this range for every intermediate coefficient can burdensome for a decoder,
  // therefore the following assertion is only enabled when configured with
  // --enable-coefficient-range-checking.
94 95
  assert(INT16_MIN <= input);
  assert(input <= INT16_MAX);
96
#endif
97 98 99 100 101 102
  return (tran_low_t)input;
}

static INLINE tran_low_t dct_const_round_shift(tran_high_t input) {
  tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
  return check_range(rv);
103
}
Yaowu Xu's avatar
Yaowu Xu committed
104

105
typedef void (*transform_1d)(const tran_low_t*, tran_low_t*);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
106 107 108 109 110

typedef struct {
  transform_1d cols, rows;  // vertical and horizontal
} transform_2d;

111
#if CONFIG_VP9_HIGHBITDEPTH
112
typedef void (*highbd_transform_1d)(const tran_low_t*, tran_low_t*, int bd);
113

114
typedef struct {
115 116
  highbd_transform_1d cols, rows;  // vertical and horizontal
} highbd_transform_2d;
117 118 119 120 121 122 123 124 125
#endif  // CONFIG_VP9_HIGHBITDEPTH

void vp9_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
                     int eob);
void vp9_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
                     int eob);
void vp9_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride,
                     int eob);
void vp9_idct16x16_add(const tran_low_t *input, uint8_t *dest, int stride, int
126
                       eob);
127
void vp9_idct32x32_add(const tran_low_t *input, uint8_t *dest, int stride,
128
                       int eob);
129

130
void vp9_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
131
                    int stride, int eob);
132
void vp9_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
133
                    int stride, int eob);
134
void vp9_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
135
                      int stride, int eob);
136

137
#if CONFIG_VP9_HIGHBITDEPTH
138
void vp9_highbd_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
139
                            int eob, int bd);
140
void vp9_highbd_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
141
                            int eob, int bd);
142 143 144 145 146 147 148 149 150
void vp9_highbd_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride,
                            int eob, int bd);
void vp9_highbd_idct16x16_add(const tran_low_t *input, uint8_t *dest,
                              int stride, int eob, int bd);
void vp9_highbd_idct32x32_add(const tran_low_t *input, uint8_t *dest,
                              int stride, int eob, int bd);
void vp9_highbd_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input,
                           uint8_t *dest, int stride, int eob, int bd);
void vp9_highbd_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input,
151
                           uint8_t *dest, int stride, int eob, int bd);
152 153
void vp9_highbd_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input,
                             uint8_t *dest, int stride, int eob, int bd);
154
#endif  // CONFIG_VP9_HIGHBITDEPTH
155 156 157 158
#ifdef __cplusplus
}  // extern "C"
#endif

Dmitry Kovalev's avatar
Dmitry Kovalev committed
159
#endif  // VP9_COMMON_VP9_IDCT_H_