/src/libvpx/vp9/common/vp9_idct.c

Source (jump to first uncovered line)
/*
 *  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.
 */

#include <math.h>

#include "./vp9_rtcd.h"
#include "./vpx_dsp_rtcd.h"
#include "vp9/common/vp9_blockd.h"
#include "vp9/common/vp9_idct.h"
#include "vpx_dsp/inv_txfm.h"
#include "vpx_ports/mem.h"

void vp9_iht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride,
                         int tx_type) {
  const transform_2d IHT_4[] = {
    { idct4_c, idct4_c },   // DCT_DCT  = 0
    { iadst4_c, idct4_c },  // ADST_DCT = 1
    { idct4_c, iadst4_c },  // DCT_ADST = 2
    { iadst4_c, iadst4_c }  // ADST_ADST = 3
  };

  int i, j;
  tran_low_t out[4 * 4];
  tran_low_t *outptr = out;
  tran_low_t temp_in[4], temp_out[4];

  // inverse transform row vectors
  for (i = 0; i < 4; ++i) {
    IHT_4[tx_type].rows(input, outptr);
    input += 4;
    outptr += 4;
  }

  // inverse transform column vectors
  for (i = 0; i < 4; ++i) {
    for (j = 0; j < 4; ++j) temp_in[j] = out[j * 4 + i];
    IHT_4[tx_type].cols(temp_in, temp_out);
    for (j = 0; j < 4; ++j) {
      dest[j * stride + i] = clip_pixel_add(dest[j * stride + i],
                                            ROUND_POWER_OF_TWO(temp_out[j], 4));
    }
  }
}

static const transform_2d IHT_8[] = {
  { idct8_c, idct8_c },   // DCT_DCT  = 0
  { iadst8_c, idct8_c },  // ADST_DCT = 1
  { idct8_c, iadst8_c },  // DCT_ADST = 2
  { iadst8_c, iadst8_c }  // ADST_ADST = 3
};

void vp9_iht8x8_64_add_c(const tran_low_t *input, uint8_t *dest, int stride,
                         int tx_type) {
  int i, j;
  tran_low_t out[8 * 8];
  tran_low_t *outptr = out;
  tran_low_t temp_in[8], temp_out[8];
  const transform_2d ht = IHT_8[tx_type];

  // inverse transform row vectors
  for (i = 0; i < 8; ++i) {
    ht.rows(input, outptr);
    input += 8;
    outptr += 8;
  }

  // inverse transform column vectors
  for (i = 0; i < 8; ++i) {
    for (j = 0; j < 8; ++j) temp_in[j] = out[j * 8 + i];
    ht.cols(temp_in, temp_out);
    for (j = 0; j < 8; ++j) {
      dest[j * stride + i] = clip_pixel_add(dest[j * stride + i],
                                            ROUND_POWER_OF_TWO(temp_out[j], 5));
    }
  }
}

static const transform_2d IHT_16[] = {
  { idct16_c, idct16_c },   // DCT_DCT  = 0
  { iadst16_c, idct16_c },  // ADST_DCT = 1
  { idct16_c, iadst16_c },  // DCT_ADST = 2
  { iadst16_c, iadst16_c }  // ADST_ADST = 3
};

void vp9_iht16x16_256_add_c(const tran_low_t *input, uint8_t *dest, int stride,
                            int tx_type) {
  int i, j;
  tran_low_t out[16 * 16];
  tran_low_t *outptr = out;
  tran_low_t temp_in[16], temp_out[16];
  const transform_2d ht = IHT_16[tx_type];

  // Rows
  for (i = 0; i < 16; ++i) {
    ht.rows(input, outptr);
    input += 16;
    outptr += 16;
  }

  // Columns
  for (i = 0; i < 16; ++i) {
    for (j = 0; j < 16; ++j) temp_in[j] = out[j * 16 + i];
    ht.cols(temp_in, temp_out);
    for (j = 0; j < 16; ++j) {
      dest[j * stride + i] = clip_pixel_add(dest[j * stride + i],
                                            ROUND_POWER_OF_TWO(temp_out[j], 6));
    }
  }
}

// idct
void vp9_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
                     int eob) {
  if (eob > 1)
    vpx_idct4x4_16_add(input, dest, stride);
  else
    vpx_idct4x4_1_add(input, dest, stride);
}

void vp9_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
                     int eob) {
  if (eob > 1)
    vpx_iwht4x4_16_add(input, dest, stride);
  else
    vpx_iwht4x4_1_add(input, dest, stride);
}

void vp9_idct8x8_add(const tran_low_t *input, uint8_t *dest, int stride,
                     int eob) {
  // If dc is 1, then input[0] is the reconstructed value, do not need
  // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.

  // The calculation can be simplified if there are not many non-zero dct
  // coefficients. Use eobs to decide what to do.
  if (eob == 1)
    // DC only DCT coefficient
    vpx_idct8x8_1_add(input, dest, stride);
  else if (eob <= 12)
    vpx_idct8x8_12_add(input, dest, stride);
  else
    vpx_idct8x8_64_add(input, dest, stride);
}

void vp9_idct16x16_add(const tran_low_t *input, uint8_t *dest, int stride,
                       int eob) {
  /* The calculation can be simplified if there are not many non-zero dct
   * coefficients. Use eobs to separate different cases. */
  if (eob == 1) /* DC only DCT coefficient. */
    vpx_idct16x16_1_add(input, dest, stride);
  else if (eob <= 10)
    vpx_idct16x16_10_add(input, dest, stride);
  else if (eob <= 38)
    vpx_idct16x16_38_add(input, dest, stride);
  else
    vpx_idct16x16_256_add(input, dest, stride);
}

void vp9_idct32x32_add(const tran_low_t *input, uint8_t *dest, int stride,
                       int eob) {
  if (eob == 1)
    vpx_idct32x32_1_add(input, dest, stride);
  else if (eob <= 34)
    // non-zero coeff only in upper-left 8x8
    vpx_idct32x32_34_add(input, dest, stride);
  else if (eob <= 135)
    // non-zero coeff only in upper-left 16x16
    vpx_idct32x32_135_add(input, dest, stride);
  else
    vpx_idct32x32_1024_add(input, dest, stride);
}

// iht
void vp9_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
                    int stride, int eob) {
  if (tx_type == DCT_DCT)
    vp9_idct4x4_add(input, dest, stride, eob);
  else
    vp9_iht4x4_16_add(input, dest, stride, tx_type);
}

void vp9_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
                    int stride, int eob) {
  if (tx_type == DCT_DCT) {
    vp9_idct8x8_add(input, dest, stride, eob);
  } else {
    vp9_iht8x8_64_add(input, dest, stride, tx_type);
  }
}

void vp9_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
                      int stride, int eob) {
  if (tx_type == DCT_DCT) {
    vp9_idct16x16_add(input, dest, stride, eob);
  } else {
    vp9_iht16x16_256_add(input, dest, stride, tx_type);
  }
}

#if CONFIG_VP9_HIGHBITDEPTH

void vp9_highbd_iht4x4_16_add_c(const tran_low_t *input, uint16_t *dest,
                                int stride, int tx_type, int bd) {
  const highbd_transform_2d IHT_4[] = {
    { vpx_highbd_idct4_c, vpx_highbd_idct4_c },   // DCT_DCT  = 0
    { vpx_highbd_iadst4_c, vpx_highbd_idct4_c },  // ADST_DCT = 1
    { vpx_highbd_idct4_c, vpx_highbd_iadst4_c },  // DCT_ADST = 2
    { vpx_highbd_iadst4_c, vpx_highbd_iadst4_c }  // ADST_ADST = 3
  };

  int i, j;
  tran_low_t out[4 * 4];
  tran_low_t *outptr = out;
  tran_low_t temp_in[4], temp_out[4];

  // Inverse transform row vectors.
  for (i = 0; i < 4; ++i) {
    IHT_4[tx_type].rows(input, outptr, bd);
    input += 4;
    outptr += 4;
  }

  // Inverse transform column vectors.
  for (i = 0; i < 4; ++i) {
    for (j = 0; j < 4; ++j) temp_in[j] = out[j * 4 + i];
    IHT_4[tx_type].cols(temp_in, temp_out, bd);
    for (j = 0; j < 4; ++j) {
      dest[j * stride + i] = highbd_clip_pixel_add(
          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 4), bd);
    }
  }
}

static const highbd_transform_2d HIGH_IHT_8[] = {
  { vpx_highbd_idct8_c, vpx_highbd_idct8_c },   // DCT_DCT  = 0
  { vpx_highbd_iadst8_c, vpx_highbd_idct8_c },  // ADST_DCT = 1
  { vpx_highbd_idct8_c, vpx_highbd_iadst8_c },  // DCT_ADST = 2
  { vpx_highbd_iadst8_c, vpx_highbd_iadst8_c }  // ADST_ADST = 3
};

void vp9_highbd_iht8x8_64_add_c(const tran_low_t *input, uint16_t *dest,
                                int stride, int tx_type, int bd) {
  int i, j;
  tran_low_t out[8 * 8];
  tran_low_t *outptr = out;
  tran_low_t temp_in[8], temp_out[8];
  const highbd_transform_2d ht = HIGH_IHT_8[tx_type];

  // Inverse transform row vectors.
  for (i = 0; i < 8; ++i) {
    ht.rows(input, outptr, bd);
    input += 8;
    outptr += 8;
  }

  // Inverse transform column vectors.
  for (i = 0; i < 8; ++i) {
    for (j = 0; j < 8; ++j) temp_in[j] = out[j * 8 + i];
    ht.cols(temp_in, temp_out, bd);
    for (j = 0; j < 8; ++j) {
      dest[j * stride + i] = highbd_clip_pixel_add(
          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd);
    }
  }
}

static const highbd_transform_2d HIGH_IHT_16[] = {
  { vpx_highbd_idct16_c, vpx_highbd_idct16_c },   // DCT_DCT  = 0
  { vpx_highbd_iadst16_c, vpx_highbd_idct16_c },  // ADST_DCT = 1
  { vpx_highbd_idct16_c, vpx_highbd_iadst16_c },  // DCT_ADST = 2
  { vpx_highbd_iadst16_c, vpx_highbd_iadst16_c }  // ADST_ADST = 3
};

void vp9_highbd_iht16x16_256_add_c(const tran_low_t *input, uint16_t *dest,
                                   int stride, int tx_type, int bd) {
  int i, j;
  tran_low_t out[16 * 16];
  tran_low_t *outptr = out;
  tran_low_t temp_in[16], temp_out[16];
  const highbd_transform_2d ht = HIGH_IHT_16[tx_type];

  // Rows
  for (i = 0; i < 16; ++i) {
    ht.rows(input, outptr, bd);
    input += 16;
    outptr += 16;
  }

  // Columns
  for (i = 0; i < 16; ++i) {
    for (j = 0; j < 16; ++j) temp_in[j] = out[j * 16 + i];
    ht.cols(temp_in, temp_out, bd);
    for (j = 0; j < 16; ++j) {
      dest[j * stride + i] = highbd_clip_pixel_add(
          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd);
    }
  }
}

// idct
void vp9_highbd_idct4x4_add(const tran_low_t *input, uint16_t *dest, int stride,
                            int eob, int bd) {
  if (eob > 1)
    vpx_highbd_idct4x4_16_add(input, dest, stride, bd);
  else
    vpx_highbd_idct4x4_1_add(input, dest, stride, bd);
}

void vp9_highbd_iwht4x4_add(const tran_low_t *input, uint16_t *dest, int stride,
                            int eob, int bd) {
  if (eob > 1)
    vpx_highbd_iwht4x4_16_add(input, dest, stride, bd);
  else
    vpx_highbd_iwht4x4_1_add(input, dest, stride, bd);
}

void vp9_highbd_idct8x8_add(const tran_low_t *input, uint16_t *dest, int stride,
                            int eob, int bd) {
  // If dc is 1, then input[0] is the reconstructed value, do not need
  // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.

  // The calculation can be simplified if there are not many non-zero dct
  // coefficients. Use eobs to decide what to do.
  // DC only DCT coefficient
  if (eob == 1) {
    vpx_highbd_idct8x8_1_add(input, dest, stride, bd);
  } else if (eob <= 12) {
    vpx_highbd_idct8x8_12_add(input, dest, stride, bd);
  } else {
    vpx_highbd_idct8x8_64_add(input, dest, stride, bd);
  }
}

void vp9_highbd_idct16x16_add(const tran_low_t *input, uint16_t *dest,
                              int stride, int eob, int bd) {
  // The calculation can be simplified if there are not many non-zero dct
  // coefficients. Use eobs to separate different cases.
  // DC only DCT coefficient.
  if (eob == 1) {
    vpx_highbd_idct16x16_1_add(input, dest, stride, bd);
  } else if (eob <= 10) {
    vpx_highbd_idct16x16_10_add(input, dest, stride, bd);
  } else if (eob <= 38) {
    vpx_highbd_idct16x16_38_add(input, dest, stride, bd);
  } else {
    vpx_highbd_idct16x16_256_add(input, dest, stride, bd);
  }
}

void vp9_highbd_idct32x32_add(const tran_low_t *input, uint16_t *dest,
                              int stride, int eob, int bd) {
  // Non-zero coeff only in upper-left 8x8
  if (eob == 1) {
    vpx_highbd_idct32x32_1_add(input, dest, stride, bd);
  } else if (eob <= 34) {
    vpx_highbd_idct32x32_34_add(input, dest, stride, bd);
  } else if (eob <= 135) {
    vpx_highbd_idct32x32_135_add(input, dest, stride, bd);
  } else {
    vpx_highbd_idct32x32_1024_add(input, dest, stride, bd);
  }
}

// iht
void vp9_highbd_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input,
                           uint16_t *dest, int stride, int eob, int bd) {
  if (tx_type == DCT_DCT)
    vp9_highbd_idct4x4_add(input, dest, stride, eob, bd);
  else
    vp9_highbd_iht4x4_16_add(input, dest, stride, tx_type, bd);
}

void vp9_highbd_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input,
                           uint16_t *dest, int stride, int eob, int bd) {
  if (tx_type == DCT_DCT) {
    vp9_highbd_idct8x8_add(input, dest, stride, eob, bd);
  } else {
    vp9_highbd_iht8x8_64_add(input, dest, stride, tx_type, bd);
  }
}

void vp9_highbd_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input,
                             uint16_t *dest, int stride, int eob, int bd) {
  if (tx_type == DCT_DCT) {
    vp9_highbd_idct16x16_add(input, dest, stride, eob, bd);
  } else {
    vp9_highbd_iht16x16_256_add(input, dest, stride, tx_type, bd);
  }
}
#endif  // CONFIG_VP9_HIGHBITDEPTH

Coverage Report

Created: 2023-03-26 07:21

Line	Count	Source (jump to first uncovered line)
1		/*
2		* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3		*
4		* Use of this source code is governed by a BSD-style license
5		* that can be found in the LICENSE file in the root of the source
6		* tree. An additional intellectual property rights grant can be found
7		* in the file PATENTS. All contributing project authors may
8		* be found in the AUTHORS file in the root of the source tree.
9		*/
10
11		#include <math.h>
12
13		#include "./vp9_rtcd.h"
14		#include "./vpx_dsp_rtcd.h"
15		#include "vp9/common/vp9_blockd.h"
16		#include "vp9/common/vp9_idct.h"
17		#include "vpx_dsp/inv_txfm.h"
18		#include "vpx_ports/mem.h"
19
20		void vp9_iht4x4_16_add_c(const tran_low_t input, uint8_t dest, int stride,
21	0	int tx_type) {
22	0	const transform_2d IHT_4[] = {
23	0	{ idct4_c, idct4_c }, // DCT_DCT = 0
24	0	{ iadst4_c, idct4_c }, // ADST_DCT = 1
25	0	{ idct4_c, iadst4_c }, // DCT_ADST = 2
26	0	{ iadst4_c, iadst4_c } // ADST_ADST = 3
27	0	};
28
29	0	int i, j;
30	0	tran_low_t out[4 * 4];
31	0	tran_low_t *outptr = out;
32	0	tran_low_t temp_in[4], temp_out[4];
33
34		// inverse transform row vectors
35	0	for (i = 0; i < 4; ++i) {
36	0	IHT_4[tx_type].rows(input, outptr);
37	0	input += 4;
38	0	outptr += 4;
39	0	}
40
41		// inverse transform column vectors
42	0	for (i = 0; i < 4; ++i) {
43	0	for (j = 0; j < 4; ++j) temp_in[j] = out[j * 4 + i];
44	0	IHT_4[tx_type].cols(temp_in, temp_out);
45	0	for (j = 0; j < 4; ++j) {
46	0	dest[j * stride + i] = clip_pixel_add(dest[j * stride + i],
47	0	ROUND_POWER_OF_TWO(temp_out[j], 4));
48	0	}
49	0	}
50	0	}
51
52		static const transform_2d IHT_8[] = {
53		{ idct8_c, idct8_c }, // DCT_DCT = 0
54		{ iadst8_c, idct8_c }, // ADST_DCT = 1
55		{ idct8_c, iadst8_c }, // DCT_ADST = 2
56		{ iadst8_c, iadst8_c } // ADST_ADST = 3
57		};
58
59		void vp9_iht8x8_64_add_c(const tran_low_t input, uint8_t dest, int stride,
60	0	int tx_type) {
61	0	int i, j;
62	0	tran_low_t out[8 * 8];
63	0	tran_low_t *outptr = out;
64	0	tran_low_t temp_in[8], temp_out[8];
65	0	const transform_2d ht = IHT_8[tx_type];
66
67		// inverse transform row vectors
68	0	for (i = 0; i < 8; ++i) {
69	0	ht.rows(input, outptr);
70	0	input += 8;
71	0	outptr += 8;
72	0	}
73
74		// inverse transform column vectors
75	0	for (i = 0; i < 8; ++i) {
76	0	for (j = 0; j < 8; ++j) temp_in[j] = out[j * 8 + i];
77	0	ht.cols(temp_in, temp_out);
78	0	for (j = 0; j < 8; ++j) {
79	0	dest[j * stride + i] = clip_pixel_add(dest[j * stride + i],
80	0	ROUND_POWER_OF_TWO(temp_out[j], 5));
81	0	}
82	0	}
83	0	}
84
85		static const transform_2d IHT_16[] = {
86		{ idct16_c, idct16_c }, // DCT_DCT = 0
87		{ iadst16_c, idct16_c }, // ADST_DCT = 1
88		{ idct16_c, iadst16_c }, // DCT_ADST = 2
89		{ iadst16_c, iadst16_c } // ADST_ADST = 3
90		};
91
92		void vp9_iht16x16_256_add_c(const tran_low_t input, uint8_t dest, int stride,
93	0	int tx_type) {
94	0	int i, j;
95	0	tran_low_t out[16 * 16];
96	0	tran_low_t *outptr = out;
97	0	tran_low_t temp_in[16], temp_out[16];
98	0	const transform_2d ht = IHT_16[tx_type];
99
100		// Rows
101	0	for (i = 0; i < 16; ++i) {
102	0	ht.rows(input, outptr);
103	0	input += 16;
104	0	outptr += 16;
105	0	}
106
107		// Columns
108	0	for (i = 0; i < 16; ++i) {
109	0	for (j = 0; j < 16; ++j) temp_in[j] = out[j * 16 + i];
110	0	ht.cols(temp_in, temp_out);
111	0	for (j = 0; j < 16; ++j) {
112	0	dest[j * stride + i] = clip_pixel_add(dest[j * stride + i],
113	0	ROUND_POWER_OF_TWO(temp_out[j], 6));
114	0	}
115	0	}
116	0	}
117
118		// idct
119		void vp9_idct4x4_add(const tran_low_t input, uint8_t dest, int stride,
120	890k	int eob) {
121	890k	if (eob > 1)
122	731k	vpx_idct4x4_16_add(input, dest, stride);
123	158k	else
124	158k	vpx_idct4x4_1_add(input, dest, stride);
125	890k	}
126
127		void vp9_iwht4x4_add(const tran_low_t input, uint8_t dest, int stride,
128	76.7k	int eob) {
129	76.7k	if (eob > 1)
130	65.2k	vpx_iwht4x4_16_add(input, dest, stride);
131	11.5k	else
132	11.5k	vpx_iwht4x4_1_add(input, dest, stride);
133	76.7k	}
134
135		void vp9_idct8x8_add(const tran_low_t input, uint8_t dest, int stride,
136	661k	int eob) {
137		// If dc is 1, then input[0] is the reconstructed value, do not need
138		// dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
139
140		// The calculation can be simplified if there are not many non-zero dct
141		// coefficients. Use eobs to decide what to do.
142	661k	if (eob == 1)
143		// DC only DCT coefficient
144	157k	vpx_idct8x8_1_add(input, dest, stride);
145	503k	else if (eob <= 12)
146	219k	vpx_idct8x8_12_add(input, dest, stride);
147	283k	else
148	283k	vpx_idct8x8_64_add(input, dest, stride);
149	661k	}
150
151		void vp9_idct16x16_add(const tran_low_t input, uint8_t dest, int stride,
152	501k	int eob) {
153		/* The calculation can be simplified if there are not many non-zero dct
154		* coefficients. Use eobs to separate different cases. */
155	501k	if (eob == 1) /* DC only DCT coefficient. */
156	165k	vpx_idct16x16_1_add(input, dest, stride);
157	335k	else if (eob <= 10)
158	124k	vpx_idct16x16_10_add(input, dest, stride);
159	210k	else if (eob <= 38)
160	79.1k	vpx_idct16x16_38_add(input, dest, stride);
161	131k	else
162	131k	vpx_idct16x16_256_add(input, dest, stride);
163	501k	}
164
165		void vp9_idct32x32_add(const tran_low_t input, uint8_t dest, int stride,
166	574k	int eob) {
167	574k	if (eob == 1)
168	213k	vpx_idct32x32_1_add(input, dest, stride);
169	361k	else if (eob <= 34)
170		// non-zero coeff only in upper-left 8x8
171	255k	vpx_idct32x32_34_add(input, dest, stride);
172	106k	else if (eob <= 135)
173		// non-zero coeff only in upper-left 16x16
174	74.3k	vpx_idct32x32_135_add(input, dest, stride);
175	31.7k	else
176	31.7k	vpx_idct32x32_1024_add(input, dest, stride);
177	574k	}
178
179		// iht
180		void vp9_iht4x4_add(TX_TYPE tx_type, const tran_low_t input, uint8_t dest,
181	470k	int stride, int eob) {
182	470k	if (tx_type == DCT_DCT)
183	266k	vp9_idct4x4_add(input, dest, stride, eob);
184	203k	else
185	203k	vp9_iht4x4_16_add(input, dest, stride, tx_type);
186	470k	}
187
188		void vp9_iht8x8_add(TX_TYPE tx_type, const tran_low_t input, uint8_t dest,
189	253k	int stride, int eob) {
190	253k	if (tx_type == DCT_DCT) {
191	140k	vp9_idct8x8_add(input, dest, stride, eob);
192	140k	} else {
193	113k	vp9_iht8x8_64_add(input, dest, stride, tx_type);
194	113k	}
195	253k	}
196
197		void vp9_iht16x16_add(TX_TYPE tx_type, const tran_low_t input, uint8_t dest,
198	205k	int stride, int eob) {
199	205k	if (tx_type == DCT_DCT) {
200	147k	vp9_idct16x16_add(input, dest, stride, eob);
201	147k	} else {
202	58.3k	vp9_iht16x16_256_add(input, dest, stride, tx_type);
203	58.3k	}
204	205k	}
205
206		#if CONFIG_VP9_HIGHBITDEPTH
207
208		void vp9_highbd_iht4x4_16_add_c(const tran_low_t input, uint16_t dest,
209	0	int stride, int tx_type, int bd) {
210	0	const highbd_transform_2d IHT_4[] = {
211	0	{ vpx_highbd_idct4_c, vpx_highbd_idct4_c }, // DCT_DCT = 0
212	0	{ vpx_highbd_iadst4_c, vpx_highbd_idct4_c }, // ADST_DCT = 1
213	0	{ vpx_highbd_idct4_c, vpx_highbd_iadst4_c }, // DCT_ADST = 2
214	0	{ vpx_highbd_iadst4_c, vpx_highbd_iadst4_c } // ADST_ADST = 3
215	0	};
216
217	0	int i, j;
218	0	tran_low_t out[4 * 4];
219	0	tran_low_t *outptr = out;
220	0	tran_low_t temp_in[4], temp_out[4];
221
222		// Inverse transform row vectors.
223	0	for (i = 0; i < 4; ++i) {
224	0	IHT_4[tx_type].rows(input, outptr, bd);
225	0	input += 4;
226	0	outptr += 4;
227	0	}
228
229		// Inverse transform column vectors.
230	0	for (i = 0; i < 4; ++i) {
231	0	for (j = 0; j < 4; ++j) temp_in[j] = out[j * 4 + i];
232	0	IHT_4[tx_type].cols(temp_in, temp_out, bd);
233	0	for (j = 0; j < 4; ++j) {
234	0	dest[j * stride + i] = highbd_clip_pixel_add(
235	0	dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 4), bd);
236	0	}
237	0	}
238	0	}
239
240		static const highbd_transform_2d HIGH_IHT_8[] = {
241		{ vpx_highbd_idct8_c, vpx_highbd_idct8_c }, // DCT_DCT = 0
242		{ vpx_highbd_iadst8_c, vpx_highbd_idct8_c }, // ADST_DCT = 1
243		{ vpx_highbd_idct8_c, vpx_highbd_iadst8_c }, // DCT_ADST = 2
244		{ vpx_highbd_iadst8_c, vpx_highbd_iadst8_c } // ADST_ADST = 3
245		};
246
247		void vp9_highbd_iht8x8_64_add_c(const tran_low_t input, uint16_t dest,
248	0	int stride, int tx_type, int bd) {
249	0	int i, j;
250	0	tran_low_t out[8 * 8];
251	0	tran_low_t *outptr = out;
252	0	tran_low_t temp_in[8], temp_out[8];
253	0	const highbd_transform_2d ht = HIGH_IHT_8[tx_type];
254
255		// Inverse transform row vectors.
256	0	for (i = 0; i < 8; ++i) {
257	0	ht.rows(input, outptr, bd);
258	0	input += 8;
259	0	outptr += 8;
260	0	}
261
262		// Inverse transform column vectors.
263	0	for (i = 0; i < 8; ++i) {
264	0	for (j = 0; j < 8; ++j) temp_in[j] = out[j * 8 + i];
265	0	ht.cols(temp_in, temp_out, bd);
266	0	for (j = 0; j < 8; ++j) {
267	0	dest[j * stride + i] = highbd_clip_pixel_add(
268	0	dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd);
269	0	}
270	0	}
271	0	}
272
273		static const highbd_transform_2d HIGH_IHT_16[] = {
274		{ vpx_highbd_idct16_c, vpx_highbd_idct16_c }, // DCT_DCT = 0
275		{ vpx_highbd_iadst16_c, vpx_highbd_idct16_c }, // ADST_DCT = 1
276		{ vpx_highbd_idct16_c, vpx_highbd_iadst16_c }, // DCT_ADST = 2
277		{ vpx_highbd_iadst16_c, vpx_highbd_iadst16_c } // ADST_ADST = 3
278		};
279
280		void vp9_highbd_iht16x16_256_add_c(const tran_low_t input, uint16_t dest,
281	0	int stride, int tx_type, int bd) {
282	0	int i, j;
283	0	tran_low_t out[16 * 16];
284	0	tran_low_t *outptr = out;
285	0	tran_low_t temp_in[16], temp_out[16];
286	0	const highbd_transform_2d ht = HIGH_IHT_16[tx_type];
287
288		// Rows
289	0	for (i = 0; i < 16; ++i) {
290	0	ht.rows(input, outptr, bd);
291	0	input += 16;
292	0	outptr += 16;
293	0	}
294
295		// Columns
296	0	for (i = 0; i < 16; ++i) {
297	0	for (j = 0; j < 16; ++j) temp_in[j] = out[j * 16 + i];
298	0	ht.cols(temp_in, temp_out, bd);
299	0	for (j = 0; j < 16; ++j) {
300	0	dest[j * stride + i] = highbd_clip_pixel_add(
301	0	dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd);
302	0	}
303	0	}
304	0	}
305
306		// idct
307		void vp9_highbd_idct4x4_add(const tran_low_t input, uint16_t dest, int stride,
308	386k	int eob, int bd) {
309	386k	if (eob > 1)
310	315k	vpx_highbd_idct4x4_16_add(input, dest, stride, bd);
311	70.8k	else
312	70.8k	vpx_highbd_idct4x4_1_add(input, dest, stride, bd);
313	386k	}
314
315		void vp9_highbd_iwht4x4_add(const tran_low_t input, uint16_t dest, int stride,
316	766k	int eob, int bd) {
317	766k	if (eob > 1)
318	646k	vpx_highbd_iwht4x4_16_add(input, dest, stride, bd);
319	120k	else
320	120k	vpx_highbd_iwht4x4_1_add(input, dest, stride, bd);
321	766k	}
322
323		void vp9_highbd_idct8x8_add(const tran_low_t input, uint16_t dest, int stride,
324	136k	int eob, int bd) {
325		// If dc is 1, then input[0] is the reconstructed value, do not need
326		// dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
327
328		// The calculation can be simplified if there are not many non-zero dct
329		// coefficients. Use eobs to decide what to do.
330		// DC only DCT coefficient
331	136k	if (eob == 1) {
332	33.1k	vpx_highbd_idct8x8_1_add(input, dest, stride, bd);
333	103k	} else if (eob <= 12) {
334	51.3k	vpx_highbd_idct8x8_12_add(input, dest, stride, bd);
335	52.4k	} else {
336	52.4k	vpx_highbd_idct8x8_64_add(input, dest, stride, bd);
337	52.4k	}
338	136k	}
339
340		void vp9_highbd_idct16x16_add(const tran_low_t input, uint16_t dest,
341	106k	int stride, int eob, int bd) {
342		// The calculation can be simplified if there are not many non-zero dct
343		// coefficients. Use eobs to separate different cases.
344		// DC only DCT coefficient.
345	106k	if (eob == 1) {
346	30.1k	vpx_highbd_idct16x16_1_add(input, dest, stride, bd);
347	76.5k	} else if (eob <= 10) {
348	34.1k	vpx_highbd_idct16x16_10_add(input, dest, stride, bd);
349	42.3k	} else if (eob <= 38) {
350	17.9k	vpx_highbd_idct16x16_38_add(input, dest, stride, bd);
351	24.4k	} else {
352	24.4k	vpx_highbd_idct16x16_256_add(input, dest, stride, bd);
353	24.4k	}
354	106k	}
355
356		void vp9_highbd_idct32x32_add(const tran_low_t input, uint16_t dest,
357	262k	int stride, int eob, int bd) {
358		// Non-zero coeff only in upper-left 8x8
359	262k	if (eob == 1) {
360	53.3k	vpx_highbd_idct32x32_1_add(input, dest, stride, bd);
361	208k	} else if (eob <= 34) {
362	147k	vpx_highbd_idct32x32_34_add(input, dest, stride, bd);
363	147k	} else if (eob <= 135) {
364	38.2k	vpx_highbd_idct32x32_135_add(input, dest, stride, bd);
365	38.2k	} else {
366	23.1k	vpx_highbd_idct32x32_1024_add(input, dest, stride, bd);
367	23.1k	}
368	262k	}
369
370		// iht
371		void vp9_highbd_iht4x4_add(TX_TYPE tx_type, const tran_low_t *input,
372	523k	uint16_t *dest, int stride, int eob, int bd) {
373	523k	if (tx_type == DCT_DCT)
374	343k	vp9_highbd_idct4x4_add(input, dest, stride, eob, bd);
375	179k	else
376	179k	vp9_highbd_iht4x4_16_add(input, dest, stride, tx_type, bd);
377	523k	}
378
379		void vp9_highbd_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input,
380	229k	uint16_t *dest, int stride, int eob, int bd) {
381	229k	if (tx_type == DCT_DCT) {
382	117k	vp9_highbd_idct8x8_add(input, dest, stride, eob, bd);
383	117k	} else {
384	111k	vp9_highbd_iht8x8_64_add(input, dest, stride, tx_type, bd);
385	111k	}
386	229k	}
387
388		void vp9_highbd_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input,
389	163k	uint16_t *dest, int stride, int eob, int bd) {
390	163k	if (tx_type == DCT_DCT) {
391	90.6k	vp9_highbd_idct16x16_add(input, dest, stride, eob, bd);
392	90.6k	} else {
393	73.1k	vp9_highbd_iht16x16_256_add(input, dest, stride, tx_type, bd);
394	73.1k	}
395	163k	}
396		#endif // CONFIG_VP9_HIGHBITDEPTH