generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Adding 8x16/16x8/32x16/16x32 transforms

Browse Source 
Adds forward, inverse transforms and scan orders.

Change-Id: Iab6994f4b0ef65e660b714d111b79b1c8172d6a8
This commit is contained in:
Debargha Mukherjee
2016-07-22 08:55:42 -07:00
parent 814986b84e
commit 7208145722
11 changed files with 2814 additions and 275 deletions
Download Patch File Download Diff File Expand all files Collapse all files

58
vp10/common/common_data.h
View File

@@ -60,41 +60,41 @@ static const uint8_t num_16x16_blocks_high_lookup[BLOCK_SIZES] = {
1, 1, 1, 1, 1, 1, 1, 2, 1, 2, 4, 2, 4, IF_EXT_PARTITION(8, 4, 8)
};
static const uint8_t num_4x4_blocks_txsize_lookup[TX_SIZES_ALL] = { 1, 4,
16, 64,
static const uint8_t num_4x4_blocks_txsize_lookup[TX_SIZES_ALL] = {
1, 4, 16, 64,
#if CONFIG_EXT_TX
2, 2
2, 2, 8, 8, 32, 32
#endif // CONFIG_EXT_TX
};
static const uint8_t num_4x4_blocks_wide_txsize_lookup[TX_SIZES_ALL] = { 1, 2,
4, 8,
static const uint8_t num_4x4_blocks_wide_txsize_lookup[TX_SIZES_ALL] = {
1, 2, 4, 8,
#if CONFIG_EXT_TX
1, 2
1, 2, 2, 4, 4, 8
#endif // CONFIG_EXT_TX
};
static const uint8_t num_4x4_blocks_high_txsize_lookup[TX_SIZES_ALL] = { 1, 2,
4, 8,
static const uint8_t num_4x4_blocks_high_txsize_lookup[TX_SIZES_ALL] = {
1, 2, 4, 8,
#if CONFIG_EXT_TX
2, 1
2, 1, 4, 2, 8, 4
#endif // CONFIG_EXT_TX
};
static const uint8_t num_4x4_blocks_txsize_log2_lookup[TX_SIZES_ALL] = { 0, 2,
4, 6,
static const uint8_t num_4x4_blocks_txsize_log2_lookup[TX_SIZES_ALL] = {
0, 2, 4, 6,
#if CONFIG_EXT_TX
1, 1
1, 1, 3, 3, 5, 5
#endif // CONFIG_EXT_TX
};
static const uint8_t num_4x4_blocks_wide_txsize_log2_lookup[TX_SIZES_ALL] = {
0, 1, 2, 3,
#if CONFIG_EXT_TX
0, 1
0, 1, 1, 2, 2, 3
#endif // CONFIG_EXT_TX
};
static const uint8_t num_4x4_blocks_high_txsize_log2_lookup[TX_SIZES_ALL] = {
0, 1, 2, 3,
#if CONFIG_EXT_TX
1, 0
1, 0, 2, 1, 3, 2
#endif // CONFIG_EXT_TX
};
@@ -374,9 +374,13 @@ static const BLOCK_SIZE txsize_to_bsize[TX_SIZES_ALL] = {
BLOCK_16X16, // TX_16X16
BLOCK_32X32, // TX_32X32
#if CONFIG_EXT_TX
BLOCK_4X8, // TX_4X8
BLOCK_8X4, // TX_8X4
#endif // CONFIG_EXT_TX
BLOCK_4X8, // TX_4X8
BLOCK_8X4, // TX_8X4
BLOCK_8X16, // TX_8X16
BLOCK_16X8, // TX_16X8
BLOCK_16X32, // TX_16X32
BLOCK_32X16, // TX_32X16
#endif // CONFIG_EXT_TX
};
static const TX_SIZE txsize_sqr_map[TX_SIZES_ALL] = {
@@ -385,9 +389,13 @@ static const TX_SIZE txsize_sqr_map[TX_SIZES_ALL] = {
TX_16X16, // TX_16X16
TX_32X32, // TX_32X32
#if CONFIG_EXT_TX
TX_4X4, // TX_4X8
TX_4X4, // TX_8X4
#endif // CONFIG_EXT_TX
TX_4X4, // TX_4X8
TX_4X4, // TX_8X4
TX_8X8, // TX_8X16
TX_8X8, // TX_16X8
TX_16X16, // TX_16X32
TX_16X16, // TX_32X16
#endif // CONFIG_EXT_TX
};
static const TX_SIZE txsize_sqr_up_map[TX_SIZES_ALL] = {
@@ -396,9 +404,13 @@ static const TX_SIZE txsize_sqr_up_map[TX_SIZES_ALL] = {
TX_16X16, // TX_16X16
TX_32X32, // TX_32X32
#if CONFIG_EXT_TX
TX_8X8, // TX_4X8
TX_8X8, // TX_8X4
#endif // CONFIG_EXT_TX
TX_8X8, // TX_4X8
TX_8X8, // TX_8X4
TX_16X16, // TX_8X16
TX_16X16, // TX_16X8
TX_32X32, // TX_16X32
TX_32X32, // TX_32X16
#endif // CONFIG_EXT_TX
};
static const TX_SIZE tx_mode_to_biggest_tx_size[TX_MODES] = {

16
vp10/common/entropy.c
View File

@@ -59,7 +59,12 @@ const uint16_t band_count_table[TX_SIZES_ALL][8] = {
{ 1, 2, 3, 4, 3, 16 - 13, 0 }, { 1, 2, 3, 4, 11, 64 - 21, 0 },
{ 1, 2, 3, 4, 11, 256 - 21, 0 }, { 1, 2, 3, 4, 11, 1024 - 21, 0 },
#if CONFIG_EXT_TX
{ 1, 2, 3, 4, 8, 32 - 18, 0 }, { 1, 2, 3, 4, 8, 32 - 18, 0 },
{ 1, 2, 3, 4, 8, 32 - 18, 0 },
{ 1, 2, 3, 4, 8, 32 - 18, 0 },
{ 1, 2, 3, 4, 11, 128 - 21, 0 },
{ 1, 2, 3, 4, 11, 128 - 21, 0 },
{ 1, 2, 3, 4, 11, 512 - 21, 0 },
{ 1, 2, 3, 4, 11, 512 - 21, 0 },
#endif // CONFIG_EXT_TX
};
@@ -67,7 +72,12 @@ const uint16_t band_cum_count_table[TX_SIZES_ALL][8] = {
{ 0, 1, 3, 6, 10, 13, 16, 0 }, { 0, 1, 3, 6, 10, 21, 64, 0 },
{ 0, 1, 3, 6, 10, 21, 256, 0 }, { 0, 1, 3, 6, 10, 21, 1024, 0 },
#if CONFIG_EXT_TX
{ 0, 1, 3, 6, 10, 18, 32, 0 }, { 0, 1, 3, 6, 10, 18, 32, 0 },
{ 0, 1, 3, 6, 10, 18, 32, 0 },
{ 0, 1, 3, 6, 10, 18, 32, 0 },
{ 0, 1, 3, 6, 10, 21, 128, 0 },
{ 0, 1, 3, 6, 10, 21, 128, 0 },
{ 0, 1, 3, 6, 10, 21, 512, 0 },
{ 0, 1, 3, 6, 10, 21, 512, 0 },
#endif // CONFIG_EXT_TX
};
@@ -116,7 +126,7 @@ const uint8_t vp10_coefband_trans_8x8plus[1024] = {
};
#if CONFIG_EXT_TX
const uint8_t vp10_coefband_trans_8x4_4x8[32] = {
const uint8_t vp10_coefband_trans_4x8_8x4[32] = {
0, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4,
4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
};

20
vp10/common/entropy.h
View File

@@ -156,7 +156,7 @@ void vp10_partial_adapt_probs(struct VP10Common *cm, int mi_row, int mi_col);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_coefband_trans_8x8plus[1024]);
#if CONFIG_EXT_TX
DECLARE_ALIGNED(16, extern const uint8_t, vp10_coefband_trans_8x4_4x8[32]);
DECLARE_ALIGNED(16, extern const uint8_t, vp10_coefband_trans_4x8_8x4[32]);
#endif // CONFIG_EXT_TX
DECLARE_ALIGNED(16, extern const uint8_t, vp10_coefband_trans_4x4[16]);
@@ -169,7 +169,7 @@ static INLINE const uint8_t *get_band_translate(TX_SIZE tx_size) {
case TX_4X4: return vp10_coefband_trans_4x4;
#if CONFIG_EXT_TX
case TX_4X8:
case TX_8X4: return vp10_coefband_trans_8x4_4x8;
return vp10_coefband_trans_4x8_8x4;
#endif // CONFIG_EXT_TX
default: return vp10_coefband_trans_8x8plus;
}
@@ -228,6 +228,22 @@ static INLINE int get_entropy_context(TX_SIZE tx_size, const ENTROPY_CONTEXT *a,
above_ec = !!*(const uint16_t *)a;
left_ec = l[0] != 0;
break;
case TX_8X16:
above_ec = !!*(const uint16_t *)a;
left_ec = !!*(const uint32_t *)l;
break;
case TX_16X8:
above_ec = !!*(const uint32_t *)a;
left_ec = !!*(const uint16_t *)l;
break;
case TX_16X32:
above_ec = !!*(const uint32_t *)a;
left_ec = !!*(const uint64_t *)l;
break;
case TX_32X16:
above_ec = !!*(const uint64_t *)a;
left_ec = !!*(const uint32_t *)l;
break;
#endif // CONFIG_EXT_TX
case TX_8X8:
above_ec = !!*(const uint16_t *)a;

10
vp10/common/enums.h
View File

@@ -138,9 +138,13 @@ typedef uint8_t TX_SIZE;
#define TX_SIZES ((TX_SIZE)4)
#if CONFIG_EXT_TX
#define TX_4X8 ((TX_SIZE)4) // 4x8 transform
#define TX_8X4 ((TX_SIZE)5) // 8x4 transform
#define TX_SIZES_ALL ((TX_SIZE)6) // Includes rectangular transforms
#define TX_4X8 ((TX_SIZE)4) // 4x8 transform
#define TX_8X4 ((TX_SIZE)5) // 8x4 transform
#define TX_8X16 ((TX_SIZE)6) // 8x16 transform
#define TX_16X8 ((TX_SIZE)7) // 16x8 transform
#define TX_16X32 ((TX_SIZE)8) // 16x32 transform
#define TX_32X16 ((TX_SIZE)9) // 32x16 transform
#define TX_SIZES_ALL ((TX_SIZE)10) // Includes rectangular transforms
#else
#define TX_SIZES_ALL ((TX_SIZE)4)
#endif // CONFIG_EXT_TX

644
vp10/common/idct.c
View File

@@ -540,6 +540,7 @@ void vp10_iht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride,
}
}
#if CONFIG_EXT_TX
void vp10_iht4x8_32_add_c(const tran_low_t *input, uint8_t *dest, int stride,
int tx_type) {
static const transform_2d IHT_4x8[] = {
@@ -547,9 +548,8 @@ void vp10_iht4x8_32_add_c(const tran_low_t *input, uint8_t *dest, int stride,
{ iadst8_c, idct4_c }, // ADST_DCT
{ idct8_c, iadst4_c }, // DCT_ADST
{ iadst8_c, iadst4_c }, // ADST_ADST
#if CONFIG_EXT_TX
{ iadst8_c, idct4_c }, // FLIPADST_DCT
{ idct8_c, iadst4_c }, // DCT_FLIPADST
{ iadst8_c, idct4_c }, // FLIPADST_DCT
{ idct8_c, iadst4_c }, // DCT_FLIPADST
{ iadst8_c, iadst4_c }, // FLIPADST_FLIPADST
{ iadst8_c, iadst4_c }, // ADST_FLIPADST
{ iadst8_c, iadst4_c }, // FLIPADST_ADST
@@ -560,34 +560,33 @@ void vp10_iht4x8_32_add_c(const tran_low_t *input, uint8_t *dest, int stride,
{ iidtx8_c, iadst4_c }, // H_ADST
{ iadst8_c, iidtx4_c }, // V_FLIPADST
{ iidtx8_c, iadst4_c }, // H_FLIPADST
#endif // CONFIG_EXT_TX
};
const int n = 4;
const int n2 = 8;
int i, j;
tran_low_t out[4][8], outtmp[4];
tran_low_t *outp = &out[0][0];
int outstride = 8;
int outstride = n2;
// inverse transform row vectors and transpose
for (i = 0; i < 8; ++i) {
for (i = 0; i < n2; ++i) {
IHT_4x8[tx_type].rows(input, outtmp);
for (j = 0; j < 4; ++j)
for (j = 0; j < n; ++j)
out[j][i] = (tran_low_t)dct_const_round_shift(outtmp[j] * Sqrt2);
input += 4;
input += n;
}
// inverse transform column vectors
for (i = 0; i < 4; ++i) {
for (i = 0; i < n; ++i) {
IHT_4x8[tx_type].cols(out[i], out[i]);
}
#if CONFIG_EXT_TX
maybe_flip_strides(&dest, &stride, &outp, &outstride, tx_type, 8, 4);
#endif
maybe_flip_strides(&dest, &stride, &outp, &outstride, tx_type, n2, n);
// Sum with the destination
for (i = 0; i < 8; ++i) {
for (j = 0; j < 4; ++j) {
for (i = 0; i < n2; ++i) {
for (j = 0; j < n; ++j) {
int d = i * stride + j;
int s = j * outstride + i;
dest[d] = clip_pixel_add(dest[d], ROUND_POWER_OF_TWO(outp[s], 5));
@@ -602,9 +601,8 @@ void vp10_iht8x4_32_add_c(const tran_low_t *input, uint8_t *dest, int stride,
{ iadst4_c, idct8_c }, // ADST_DCT
{ idct4_c, iadst8_c }, // DCT_ADST
{ iadst4_c, iadst8_c }, // ADST_ADST
#if CONFIG_EXT_TX
{ iadst4_c, idct8_c }, // FLIPADST_DCT
{ idct4_c, iadst8_c }, // DCT_FLIPADST
{ iadst4_c, idct8_c }, // FLIPADST_DCT
{ idct4_c, iadst8_c }, // DCT_FLIPADST
{ iadst4_c, iadst8_c }, // FLIPADST_FLIPADST
{ iadst4_c, iadst8_c }, // ADST_FLIPADST
{ iadst4_c, iadst8_c }, // FLIPADST_ADST
@@ -615,34 +613,33 @@ void vp10_iht8x4_32_add_c(const tran_low_t *input, uint8_t *dest, int stride,
{ iidtx4_c, iadst8_c }, // H_ADST
{ iadst4_c, iidtx8_c }, // V_FLIPADST
{ iidtx4_c, iadst8_c }, // H_FLIPADST
#endif // CONFIG_EXT_TX
};
const int n = 4;
const int n2 = 8;
int i, j;
tran_low_t out[8][4], outtmp[8];
tran_low_t *outp = &out[0][0];
int outstride = 4;
int outstride = n;
// inverse transform row vectors and transpose
for (i = 0; i < 4; ++i) {
for (i = 0; i < n; ++i) {
IHT_8x4[tx_type].rows(input, outtmp);
for (j = 0; j < 8; ++j)
for (j = 0; j < n2; ++j)
out[j][i] = (tran_low_t)dct_const_round_shift(outtmp[j] * Sqrt2);
input += 8;
input += n2;
}
// inverse transform column vectors
for (i = 0; i < 8; ++i) {
for (i = 0; i < n2; ++i) {
IHT_8x4[tx_type].cols(out[i], out[i]);
}
#if CONFIG_EXT_TX
maybe_flip_strides(&dest, &stride, &outp, &outstride, tx_type, 4, 8);
#endif
maybe_flip_strides(&dest, &stride, &outp, &outstride, tx_type, n, n2);
// Sum with the destination
for (i = 0; i < 4; ++i) {
for (j = 0; j < 8; ++j) {
for (i = 0; i < n; ++i) {
for (j = 0; j < n2; ++j) {
int d = i * stride + j;
int s = j * outstride + i;
dest[d] = clip_pixel_add(dest[d], ROUND_POWER_OF_TWO(outp[s], 5));
@@ -650,6 +647,219 @@ void vp10_iht8x4_32_add_c(const tran_low_t *input, uint8_t *dest, int stride,
}
}
void vp10_iht8x16_128_add_c(const tran_low_t *input, uint8_t *dest,
int stride, int tx_type) {
static const transform_2d IHT_8x16[] = {
{ idct16_c, idct8_c }, // DCT_DCT
{ iadst16_c, idct8_c }, // ADST_DCT
{ idct16_c, iadst8_c }, // DCT_ADST
{ iadst16_c, iadst8_c }, // ADST_ADST
{ iadst16_c, idct8_c }, // FLIPADST_DCT
{ idct16_c, iadst8_c }, // DCT_FLIPADST
{ iadst16_c, iadst8_c }, // FLIPADST_FLIPADST
{ iadst16_c, iadst8_c }, // ADST_FLIPADST
{ iadst16_c, iadst8_c }, // FLIPADST_ADST
{ iidtx16_c, iidtx8_c }, // IDTX
{ idct16_c, iidtx8_c }, // V_DCT
{ iidtx16_c, idct8_c }, // H_DCT
{ iadst16_c, iidtx8_c }, // V_ADST
{ iidtx16_c, iadst8_c }, // H_ADST
{ iadst16_c, iidtx8_c }, // V_FLIPADST
{ iidtx16_c, iadst8_c }, // H_FLIPADST
};
const int n = 8;
const int n2 = 16;
int i, j;
tran_low_t out[8][16], outtmp[8];
tran_low_t *outp = &out[0][0];
int outstride = n2;
// inverse transform row vectors and transpose
for (i = 0; i < n2; ++i) {
IHT_8x16[tx_type].rows(input, outtmp);
for (j = 0; j < n; ++j)
out[j][i] = (tran_low_t)dct_const_round_shift(outtmp[j] * Sqrt2);
input += n;
}
// inverse transform column vectors
for (i = 0; i < n; ++i) {
IHT_8x16[tx_type].cols(out[i], out[i]);
}
maybe_flip_strides(&dest, &stride, &outp, &outstride, tx_type, n2, n);
// Sum with the destination
for (i = 0; i < n2; ++i) {
for (j = 0; j < n; ++j) {
int d = i * stride + j;
int s = j * outstride + i;
dest[d] = clip_pixel_add(dest[d], ROUND_POWER_OF_TWO(outp[s], 6));
}
}
}
void vp10_iht16x8_128_add_c(const tran_low_t *input, uint8_t *dest,
int stride, int tx_type) {
static const transform_2d IHT_16x8[] = {
{ idct8_c, idct16_c }, // DCT_DCT
{ iadst8_c, idct16_c }, // ADST_DCT
{ idct8_c, iadst16_c }, // DCT_ADST
{ iadst8_c, iadst16_c }, // ADST_ADST
{ iadst8_c, idct16_c }, // FLIPADST_DCT
{ idct8_c, iadst16_c }, // DCT_FLIPADST
{ iadst8_c, iadst16_c }, // FLIPADST_FLIPADST
{ iadst8_c, iadst16_c }, // ADST_FLIPADST
{ iadst8_c, iadst16_c }, // FLIPADST_ADST
{ iidtx8_c, iidtx16_c }, // IDTX
{ idct8_c, iidtx16_c }, // V_DCT
{ iidtx8_c, idct16_c }, // H_DCT
{ iadst8_c, iidtx16_c }, // V_ADST
{ iidtx8_c, iadst16_c }, // H_ADST
{ iadst8_c, iidtx16_c }, // V_FLIPADST
{ iidtx8_c, iadst16_c }, // H_FLIPADST
};
const int n = 8;
const int n2 = 16;
int i, j;
tran_low_t out[16][8], outtmp[16];
tran_low_t *outp = &out[0][0];
int outstride = n;
// inverse transform row vectors and transpose
for (i = 0; i < n; ++i) {
IHT_16x8[tx_type].rows(input, outtmp);
for (j = 0; j < n2; ++j)
out[j][i] = (tran_low_t)dct_const_round_shift(outtmp[j] * Sqrt2);
input += n2;
}
// inverse transform column vectors
for (i = 0; i < n2; ++i) {
IHT_16x8[tx_type].cols(out[i], out[i]);
}
maybe_flip_strides(&dest, &stride, &outp, &outstride, tx_type, n, n2);
// Sum with the destination
for (i = 0; i < n; ++i) {
for (j = 0; j < n2; ++j) {
int d = i * stride + j;
int s = j * outstride + i;
dest[d] = clip_pixel_add(dest[d], ROUND_POWER_OF_TWO(outp[s], 6));
}
}
}
void vp10_iht16x32_512_add_c(const tran_low_t *input, uint8_t *dest,
int stride, int tx_type) {
static const transform_2d IHT_16x32[] = {
{ idct32_c, idct16_c }, // DCT_DCT
{ ihalfright32_c, idct16_c }, // ADST_DCT
{ idct32_c, iadst16_c }, // DCT_ADST
{ ihalfright32_c, iadst16_c }, // ADST_ADST
{ ihalfright32_c, idct16_c }, // FLIPADST_DCT
{ idct32_c, iadst16_c }, // DCT_FLIPADST
{ ihalfright32_c, iadst16_c }, // FLIPADST_FLIPADST
{ ihalfright32_c, iadst16_c }, // ADST_FLIPADST
{ ihalfright32_c, iadst16_c }, // FLIPADST_ADST
{ iidtx32_c, iidtx16_c }, // IDTX
{ idct32_c, iidtx16_c }, // V_DCT
{ iidtx32_c, idct16_c }, // H_DCT
{ ihalfright32_c, iidtx16_c }, // V_ADST
{ iidtx32_c, iadst16_c }, // H_ADST
{ ihalfright32_c, iidtx16_c }, // V_FLIPADST
{ iidtx32_c, iadst16_c }, // H_FLIPADST
};
const int n = 16;
const int n2 = 32;
int i, j;
tran_low_t out[16][32], outtmp[16];
tran_low_t *outp = &out[0][0];
int outstride = n2;
// inverse transform row vectors and transpose
for (i = 0; i < n2; ++i) {
IHT_16x32[tx_type].rows(input, outtmp);
for (j = 0; j < n; ++j)
out[j][i] = (tran_low_t)dct_const_round_shift(outtmp[j] * Sqrt2);
input += n;
}
// inverse transform column vectors
for (i = 0; i < n; ++i) {
IHT_16x32[tx_type].cols(out[i], out[i]);
}
maybe_flip_strides(&dest, &stride, &outp, &outstride, tx_type, n2, n);
// Sum with the destination
for (i = 0; i < n2; ++i) {
for (j = 0; j < n; ++j) {
int d = i * stride + j;
int s = j * outstride + i;
dest[d] = clip_pixel_add(dest[d], ROUND_POWER_OF_TWO(outp[s], 6));
}
}
}
void vp10_iht32x16_512_add_c(const tran_low_t *input, uint8_t *dest,
int stride, int tx_type) {
static const transform_2d IHT_32x16[] = {
{ idct16_c, idct32_c }, // DCT_DCT
{ iadst16_c, idct32_c }, // ADST_DCT
{ idct16_c, ihalfright32_c }, // DCT_ADST
{ iadst16_c, ihalfright32_c }, // ADST_ADST
{ iadst16_c, idct32_c }, // FLIPADST_DCT
{ idct16_c, ihalfright32_c }, // DCT_FLIPADST
{ iadst16_c, ihalfright32_c }, // FLIPADST_FLIPADST
{ iadst16_c, ihalfright32_c }, // ADST_FLIPADST
{ iadst16_c, ihalfright32_c }, // FLIPADST_ADST
{ iidtx16_c, iidtx32_c }, // IDTX
{ idct16_c, iidtx32_c }, // V_DCT
{ iidtx16_c, idct32_c }, // H_DCT
{ iadst16_c, iidtx32_c }, // V_ADST
{ iidtx16_c, ihalfright32_c }, // H_ADST
{ iadst16_c, iidtx32_c }, // V_FLIPADST
{ iidtx16_c, ihalfright32_c }, // H_FLIPADST
};
const int n = 16;
const int n2 = 32;
int i, j;
tran_low_t out[32][16], outtmp[32];
tran_low_t *outp = &out[0][0];
int outstride = n;
// inverse transform row vectors and transpose
for (i = 0; i < n; ++i) {
IHT_32x16[tx_type].rows(input, outtmp);
for (j = 0; j < n2; ++j)
out[j][i] = (tran_low_t)dct_const_round_shift(outtmp[j] * Sqrt2);
input += n2;
}
// inverse transform column vectors
for (i = 0; i < n2; ++i) {
IHT_32x16[tx_type].cols(out[i], out[i]);
}
maybe_flip_strides(&dest, &stride, &outp, &outstride, tx_type, n, n2);
// Sum with the destination
for (i = 0; i < n; ++i) {
for (j = 0; j < n2; ++j) {
int d = i * stride + j;
int s = j * outstride + i;
dest[d] = clip_pixel_add(dest[d], ROUND_POWER_OF_TWO(outp[s], 6));
}
}
}
#endif // CONFIG_EXT_TX
void vp10_iht8x8_64_add_c(const tran_low_t *input, uint8_t *dest, int stride,
int tx_type) {
static const transform_2d IHT_8[] = {
@@ -930,16 +1140,40 @@ void vp10_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest, int stride,
}
#if CONFIG_EXT_TX
void vp10_inv_txfm_add_8x4(const tran_low_t *input, uint8_t *dest, int stride,
int eob, TX_TYPE tx_type) {
(void)eob;
void vp10_inv_txfm_add_4x8(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type) {
(void) eob;
vp10_iht4x8_32_add(input, dest, stride, tx_type);
}
void vp10_inv_txfm_add_8x4(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type) {
(void) eob;
vp10_iht8x4_32_add(input, dest, stride, tx_type);
}
void vp10_inv_txfm_add_4x8(const tran_low_t *input, uint8_t *dest, int stride,
int eob, TX_TYPE tx_type) {
(void)eob;
vp10_iht4x8_32_add(input, dest, stride, tx_type);
void vp10_inv_txfm_add_8x16(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type) {
(void) eob;
vp10_iht8x16_128_add(input, dest, stride, tx_type);
}
void vp10_inv_txfm_add_16x8(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type) {
(void) eob;
vp10_iht16x8_128_add(input, dest, stride, tx_type);
}
void vp10_inv_txfm_add_16x32(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type) {
(void) eob;
vp10_iht16x32_512_add(input, dest, stride, tx_type);
}
void vp10_inv_txfm_add_32x16(const tran_low_t *input, uint8_t *dest,
int stride, int eob, TX_TYPE tx_type) {
(void) eob;
vp10_iht32x16_512_add(input, dest, stride, tx_type);
}
#endif // CONFIG_EXT_TX
@@ -1116,34 +1350,36 @@ void vp10_highbd_iht4x8_32_add_c(const tran_low_t *input, uint8_t *dest8,
{ vpx_highbd_iadst8_c, highbd_iidtx4_c }, // V_FLIPADST
{ highbd_iidtx8_c, vpx_highbd_iadst4_c }, // H_FLIPADST
};
const int n = 4;
const int n2 = 8;
uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
int i, j;
tran_low_t out[4][8], outtmp[4];
tran_low_t *outp = &out[0][0];
int outstride = 8;
int outstride = n2;
// inverse transform row vectors, and transpose
for (i = 0; i < 8; ++i) {
for (i = 0; i < n2; ++i) {
HIGH_IHT_4x8[tx_type].rows(input, outtmp, bd);
for (j = 0; j < 4; ++j) {
out[j][i] =
HIGHBD_WRAPLOW(highbd_dct_const_round_shift(outtmp[j] * Sqrt2), bd);
for (j = 0; j < n; ++j) {
out[j][i] = HIGHBD_WRAPLOW(
highbd_dct_const_round_shift(outtmp[j] * Sqrt2), bd);
}
input += 4;
input += n;
}
// inverse transform column vectors
for (i = 0; i < 4; ++i) {
for (i = 0; i < n; ++i) {
HIGH_IHT_4x8[tx_type].cols(out[i], out[i], bd);
}
maybe_flip_strides16(&dest, &stride, &outp, &outstride, tx_type, 8, 4);
maybe_flip_strides16(&dest, &stride, &outp, &outstride, tx_type, n2, n);
// Sum with the destination
for (i = 0; i < 8; ++i) {
for (j = 0; j < 4; ++j) {
for (i = 0; i < n2; ++i) {
for (j = 0; j < n; ++j) {
int d = i * stride + j;
int s = j * outstride + i;
dest[d] =
@@ -1172,34 +1408,36 @@ void vp10_highbd_iht8x4_32_add_c(const tran_low_t *input, uint8_t *dest8,
{ vpx_highbd_iadst4_c, highbd_iidtx8_c }, // V_FLIPADST
{ highbd_iidtx4_c, vpx_highbd_iadst8_c }, // H_FLIPADST
};
const int n = 4;
const int n2 = 8;
uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
int i, j;
tran_low_t out[8][4], outtmp[8];
tran_low_t *outp = &out[0][0];
int outstride = 4;
int outstride = n;
// inverse transform row vectors, and transpose
for (i = 0; i < 4; ++i) {
for (i = 0; i < n; ++i) {
HIGH_IHT_8x4[tx_type].rows(input, outtmp, bd);
for (j = 0; j < 8; ++j) {
out[j][i] =
HIGHBD_WRAPLOW(highbd_dct_const_round_shift(outtmp[j] * Sqrt2), bd);
for (j = 0; j < n2; ++j) {
out[j][i] = HIGHBD_WRAPLOW(
highbd_dct_const_round_shift(outtmp[j] * Sqrt2), bd);
}
input += 8;
input += n2;
}
// inverse transform column vectors
for (i = 0; i < 8; ++i) {
for (i = 0; i < n2; ++i) {
HIGH_IHT_8x4[tx_type].cols(out[i], out[i], bd);
}
maybe_flip_strides16(&dest, &stride, &outp, &outstride, tx_type, 4, 8);
maybe_flip_strides16(&dest, &stride, &outp, &outstride, tx_type, n, n2);
// Sum with the destination
for (i = 0; i < 4; ++i) {
for (j = 0; j < 8; ++j) {
for (i = 0; i < n; ++i) {
for (j = 0; j < n2; ++j) {
int d = i * stride + j;
int s = j * outstride + i;
dest[d] =
@@ -1207,6 +1445,234 @@ void vp10_highbd_iht8x4_32_add_c(const tran_low_t *input, uint8_t *dest8,
}
}
}
void vp10_highbd_iht8x16_128_add_c(const tran_low_t *input, uint8_t *dest8,
int stride, int tx_type, int bd) {
static const highbd_transform_2d HIGH_IHT_8x16[] = {
{ vpx_highbd_idct16_c, vpx_highbd_idct8_c }, // DCT_DCT
{ vpx_highbd_iadst16_c, vpx_highbd_idct8_c }, // ADST_DCT
{ vpx_highbd_idct16_c, vpx_highbd_iadst8_c }, // DCT_ADST
{ vpx_highbd_iadst16_c, vpx_highbd_iadst8_c }, // ADST_ADST
{ vpx_highbd_iadst16_c, vpx_highbd_idct8_c }, // FLIPADST_DCT
{ vpx_highbd_idct16_c, vpx_highbd_iadst8_c }, // DCT_FLIPADST
{ vpx_highbd_iadst16_c, vpx_highbd_iadst8_c }, // FLIPADST_FLIPADST
{ vpx_highbd_iadst16_c, vpx_highbd_iadst8_c }, // ADST_FLIPADST
{ vpx_highbd_iadst16_c, vpx_highbd_iadst8_c }, // FLIPADST_ADST
{ highbd_iidtx16_c, highbd_iidtx8_c }, // IDTX
{ vpx_highbd_idct16_c, highbd_iidtx8_c }, // V_DCT
{ highbd_iidtx16_c, vpx_highbd_idct8_c }, // H_DCT
{ vpx_highbd_iadst16_c, highbd_iidtx8_c }, // V_ADST
{ highbd_iidtx16_c, vpx_highbd_iadst8_c }, // H_ADST
{ vpx_highbd_iadst16_c, highbd_iidtx8_c }, // V_FLIPADST
{ highbd_iidtx16_c, vpx_highbd_iadst8_c }, // H_FLIPADST
};
const int n = 8;
const int n2 = 16;
uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
int i, j;
tran_low_t out[8][16], outtmp[8];
tran_low_t *outp = &out[0][0];
int outstride = n2;
// inverse transform row vectors, and transpose
for (i = 0; i < n2; ++i) {
HIGH_IHT_8x16[tx_type].rows(input, outtmp, bd);
for (j = 0; j < n; ++j)
out[j][i] = HIGHBD_WRAPLOW(
highbd_dct_const_round_shift(outtmp[j] * Sqrt2), bd);
input += n;
}
// inverse transform column vectors
for (i = 0; i < n; ++i) {
HIGH_IHT_8x16[tx_type].cols(out[i], out[i], bd);
}
maybe_flip_strides16(&dest, &stride, &outp, &outstride, tx_type, n2, n);
// Sum with the destination
for (i = 0; i < n2; ++i) {
for (j = 0; j < n; ++j) {
int d = i * stride + j;
int s = j * outstride + i;
dest[d] = highbd_clip_pixel_add(dest[d],
ROUND_POWER_OF_TWO(outp[s], 6), bd);
}
}
}
void vp10_highbd_iht16x8_128_add_c(const tran_low_t *input, uint8_t *dest8,
int stride, int tx_type, int bd) {
static const highbd_transform_2d HIGH_IHT_16x8[] = {
{ vpx_highbd_idct8_c, vpx_highbd_idct16_c }, // DCT_DCT
{ vpx_highbd_iadst8_c, vpx_highbd_idct16_c }, // ADST_DCT
{ vpx_highbd_idct8_c, vpx_highbd_iadst16_c }, // DCT_ADST
{ vpx_highbd_iadst8_c, vpx_highbd_iadst16_c }, // ADST_ADST
{ vpx_highbd_iadst8_c, vpx_highbd_idct16_c }, // FLIPADST_DCT
{ vpx_highbd_idct8_c, vpx_highbd_iadst16_c }, // DCT_FLIPADST
{ vpx_highbd_iadst8_c, vpx_highbd_iadst16_c }, // FLIPADST_FLIPADST
{ vpx_highbd_iadst8_c, vpx_highbd_iadst16_c }, // ADST_FLIPADST
{ vpx_highbd_iadst8_c, vpx_highbd_iadst16_c }, // FLIPADST_ADST
{ highbd_iidtx8_c, highbd_iidtx16_c }, // IDTX
{ vpx_highbd_idct8_c, highbd_iidtx16_c }, // V_DCT
{ highbd_iidtx8_c, vpx_highbd_idct16_c }, // H_DCT
{ vpx_highbd_iadst8_c, highbd_iidtx16_c }, // V_ADST
{ highbd_iidtx8_c, vpx_highbd_iadst16_c }, // H_ADST
{ vpx_highbd_iadst8_c, highbd_iidtx16_c }, // V_FLIPADST
{ highbd_iidtx8_c, vpx_highbd_iadst16_c }, // H_FLIPADST
};
const int n = 8;
const int n2 = 16;
uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
int i, j;
tran_low_t out[16][8], outtmp[16];
tran_low_t *outp = &out[0][0];
int outstride = n;
// inverse transform row vectors, and transpose
for (i = 0; i < n; ++i) {
HIGH_IHT_16x8[tx_type].rows(input, outtmp, bd);
for (j = 0; j < n2; ++j)
out[j][i] = HIGHBD_WRAPLOW(
highbd_dct_const_round_shift(outtmp[j] * Sqrt2), bd);
input += n2;
}
// inverse transform column vectors
for (i = 0; i < n2; ++i) {
HIGH_IHT_16x8[tx_type].cols(out[i], out[i], bd);
}
maybe_flip_strides16(&dest, &stride, &outp, &outstride, tx_type, n, n2);
// Sum with the destination
for (i = 0; i < n; ++i) {
for (j = 0; j < n2; ++j) {
int d = i * stride + j;
int s = j * outstride + i;
dest[d] = highbd_clip_pixel_add(dest[d],
ROUND_POWER_OF_TWO(outp[s], 6), bd);
}
}
}
void vp10_highbd_iht16x32_512_add_c(const tran_low_t *input, uint8_t *dest8,
int stride, int tx_type, int bd) {
static const highbd_transform_2d HIGH_IHT_16x32[] = {
{ vpx_highbd_idct32_c, vpx_highbd_idct16_c }, // DCT_DCT
{ highbd_ihalfright32_c, vpx_highbd_idct16_c }, // ADST_DCT
{ vpx_highbd_idct32_c, vpx_highbd_iadst16_c }, // DCT_ADST
{ highbd_ihalfright32_c, vpx_highbd_iadst16_c }, // ADST_ADST
{ highbd_ihalfright32_c, vpx_highbd_idct16_c }, // FLIPADST_DCT
{ vpx_highbd_idct32_c, vpx_highbd_iadst16_c }, // DCT_FLIPADST
{ highbd_ihalfright32_c, vpx_highbd_iadst16_c }, // FLIPADST_FLIPADST
{ highbd_ihalfright32_c, vpx_highbd_iadst16_c }, // ADST_FLIPADST
{ highbd_ihalfright32_c, vpx_highbd_iadst16_c }, // FLIPADST_ADST
{ highbd_iidtx32_c, highbd_iidtx16_c }, // IDTX
{ vpx_highbd_idct32_c, highbd_iidtx16_c }, // V_DCT
{ highbd_iidtx32_c, vpx_highbd_idct16_c }, // H_DCT
{ highbd_ihalfright32_c, highbd_iidtx16_c }, // V_ADST
{ highbd_iidtx32_c, vpx_highbd_iadst16_c }, // H_ADST
{ highbd_ihalfright32_c, highbd_iidtx16_c }, // V_FLIPADST
{ highbd_iidtx32_c, vpx_highbd_iadst16_c }, // H_FLIPADST
};
const int n = 16;
const int n2 = 32;
uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
int i, j;
tran_low_t out[16][32], outtmp[16];
tran_low_t *outp = &out[0][0];
int outstride = n2;
// inverse transform row vectors, and transpose
for (i = 0; i < n2; ++i) {
HIGH_IHT_16x32[tx_type].rows(input, outtmp, bd);
for (j = 0; j < n; ++j)
out[j][i] = HIGHBD_WRAPLOW(
highbd_dct_const_round_shift(outtmp[j] * Sqrt2), bd);
input += n;
}
// inverse transform column vectors
for (i = 0; i < n; ++i) {
HIGH_IHT_16x32[tx_type].cols(out[i], out[i], bd);
}
maybe_flip_strides16(&dest, &stride, &outp, &outstride, tx_type, n2, n);
// Sum with the destination
for (i = 0; i < n2; ++i) {
for (j = 0; j < n; ++j) {
int d = i * stride + j;
int s = j * outstride + i;
dest[d] = highbd_clip_pixel_add(dest[d],
ROUND_POWER_OF_TWO(outp[s], 6), bd);
}
}
}
void vp10_highbd_iht32x16_512_add_c(const tran_low_t *input, uint8_t *dest8,
int stride, int tx_type, int bd) {
static const highbd_transform_2d HIGH_IHT_32x16[] = {
{ vpx_highbd_idct16_c, vpx_highbd_idct32_c }, // DCT_DCT
{ vpx_highbd_iadst16_c, vpx_highbd_idct32_c }, // ADST_DCT
{ vpx_highbd_idct16_c, highbd_ihalfright32_c }, // DCT_ADST
{ vpx_highbd_iadst16_c, highbd_ihalfright32_c }, // ADST_ADST
{ vpx_highbd_iadst16_c, vpx_highbd_idct32_c }, // FLIPADST_DCT
{ vpx_highbd_idct16_c, highbd_ihalfright32_c }, // DCT_FLIPADST
{ vpx_highbd_iadst16_c, highbd_ihalfright32_c }, // FLIPADST_FLIPADST
{ vpx_highbd_iadst16_c, highbd_ihalfright32_c }, // ADST_FLIPADST
{ vpx_highbd_iadst16_c, highbd_ihalfright32_c }, // FLIPADST_ADST
{ highbd_iidtx16_c, highbd_iidtx32_c }, // IDTX
{ vpx_highbd_idct16_c, highbd_iidtx32_c }, // V_DCT
{ highbd_iidtx16_c, vpx_highbd_idct32_c }, // H_DCT
{ vpx_highbd_iadst16_c, highbd_iidtx32_c }, // V_ADST
{ highbd_iidtx16_c, highbd_ihalfright32_c }, // H_ADST
{ vpx_highbd_iadst16_c, highbd_iidtx32_c }, // V_FLIPADST
{ highbd_iidtx16_c, highbd_ihalfright32_c }, // H_FLIPADST
};
const int n = 16;
const int n2 = 32;
uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
int i, j;
tran_low_t out[32][16], outtmp[32];
tran_low_t *outp = &out[0][0];
int outstride = n;
// inverse transform row vectors, and transpose
for (i = 0; i < n; ++i) {
HIGH_IHT_32x16[tx_type].rows(input, outtmp, bd);
for (j = 0; j < n2; ++j)
out[j][i] = HIGHBD_WRAPLOW(
highbd_dct_const_round_shift(outtmp[j] * Sqrt2), bd);
input += n2;
}
// inverse transform column vectors
for (i = 0; i < n2; ++i) {
HIGH_IHT_32x16[tx_type].cols(out[i], out[i], bd);
}
maybe_flip_strides16(&dest, &stride, &outp, &outstride, tx_type, n, n2);
// Sum with the destination
for (i = 0; i < n; ++i) {
for (j = 0; j < n2; ++j) {
int d = i * stride + j;
int s = j * outstride + i;
dest[d] = highbd_clip_pixel_add(dest[d],
ROUND_POWER_OF_TWO(outp[s], 6), bd);
}
}
}
#endif // CONFIG_EXT_TX
void vp10_highbd_iht8x8_64_add_c(const tran_low_t *input, uint8_t *dest8,
@@ -1511,19 +1977,47 @@ void vp10_highbd_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest,
}
#if CONFIG_EXT_TX
void vp10_highbd_inv_txfm_add_8x4(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd,
TX_TYPE tx_type) {
(void)eob;
vp10_highbd_iht8x4_32_add_c(input, dest, stride, tx_type, bd);
}
void vp10_highbd_inv_txfm_add_4x8(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd,
TX_TYPE tx_type) {
(void)eob;
(void) eob;
vp10_highbd_iht4x8_32_add_c(input, dest, stride, tx_type, bd);
}
void vp10_highbd_inv_txfm_add_8x4(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd,
TX_TYPE tx_type) {
(void) eob;
vp10_highbd_iht8x4_32_add_c(input, dest, stride, tx_type, bd);
}
void vp10_highbd_inv_txfm_add_8x16(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd,
TX_TYPE tx_type) {
(void) eob;
vp10_highbd_iht8x16_128_add_c(input, dest, stride, tx_type, bd);
}
void vp10_highbd_inv_txfm_add_16x8(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd,
TX_TYPE tx_type) {
(void) eob;
vp10_highbd_iht16x8_128_add_c(input, dest, stride, tx_type, bd);
}
void vp10_highbd_inv_txfm_add_16x32(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd,
TX_TYPE tx_type) {
(void) eob;
vp10_highbd_iht16x32_512_add_c(input, dest, stride, tx_type, bd);
}
void vp10_highbd_inv_txfm_add_32x16(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd,
TX_TYPE tx_type) {
(void) eob;
vp10_highbd_iht32x16_512_add_c(input, dest, stride, tx_type, bd);
}
#endif // CONFIG_EXT_TX
void vp10_highbd_inv_txfm_add_8x8(const tran_low_t *input, uint8_t *dest,
@@ -1661,6 +2155,18 @@ void inv_txfm_add(const tran_low_t *input, uint8_t *dest, int stride,
case TX_8X4:
vp10_inv_txfm_add_8x4(input, dest, stride, eob, tx_type);
break;
case TX_8X16:
vp10_inv_txfm_add_8x16(input, dest, stride, eob, tx_type);
break;
case TX_16X8:
vp10_inv_txfm_add_16x8(input, dest, stride, eob, tx_type);
break;
case TX_16X32:
vp10_inv_txfm_add_16x32(input, dest, stride, eob, tx_type);
break;
case TX_32X16:
vp10_inv_txfm_add_32x16(input, dest, stride, eob, tx_type);
break;
#endif // CONFIG_EXT_TX
case TX_4X4:
// this is like vp10_short_idct4x4 but has a special case around eob<=1
@@ -1698,6 +2204,18 @@ void highbd_inv_txfm_add(const tran_low_t *input, uint8_t *dest, int stride,
case TX_8X4:
vp10_highbd_inv_txfm_add_8x4(input, dest, stride, eob, bd, tx_type);
break;
case TX_8X16:
vp10_highbd_inv_txfm_add_8x16(input, dest, stride, eob, bd, tx_type);
break;
case TX_16X8:
vp10_highbd_inv_txfm_add_16x8(input, dest, stride, eob, bd, tx_type);
break;
case TX_16X32:
vp10_highbd_inv_txfm_add_16x32(input, dest, stride, eob, bd, tx_type);
break;
case TX_32X16:
vp10_highbd_inv_txfm_add_32x16(input, dest, stride, eob, bd, tx_type);
break;
#endif // CONFIG_EXT_TX
case TX_4X4:
// this is like vp10_short_idct4x4 but has a special case around eob<=1

1765
vp10/common/scan.c
View File

File diff suppressed because it is too large Load Diff

128
vp10/common/vp10_rtcd_defs.pl
View File

@@ -60,11 +60,23 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x4_16_add/;
add_proto qw/void vp10_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x8_32_add/;
add_proto qw/void vp10_iht8x4_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x4_32_add/;
add_proto qw/void vp10_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x8_32_add/;
add_proto qw/void vp10_iht8x16_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x16_128_add/;
add_proto qw/void vp10_iht16x8_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht16x8_128_add/;
add_proto qw/void vp10_iht16x32_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht16x32_512_add/;
add_proto qw/void vp10_iht32x16_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht32x16_512_add/;
add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x8_64_add/;
@@ -126,11 +138,23 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x4_16_add sse2/;
add_proto qw/void vp10_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x8_32_add/;
add_proto qw/void vp10_iht8x4_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x4_32_add/;
add_proto qw/void vp10_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x8_32_add/;
add_proto qw/void vp10_iht8x16_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x16_128_add/;
add_proto qw/void vp10_iht16x8_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht16x8_128_add/;
add_proto qw/void vp10_iht16x32_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht16x32_512_add/;
add_proto qw/void vp10_iht32x16_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht32x16_512_add/;
add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x8_64_add sse2/;
@@ -195,11 +219,23 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x4_16_add/;
add_proto qw/void vp10_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x8_32_add/;
add_proto qw/void vp10_iht8x4_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x4_32_add/;
add_proto qw/void vp10_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x8_32_add/;
add_proto qw/void vp10_iht8x16_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x16_128_add/;
add_proto qw/void vp10_iht16x8_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht16x8_128_add/;
add_proto qw/void vp10_iht16x32_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht16x32_512_add/;
add_proto qw/void vp10_iht32x16_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht32x16_512_add/;
add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x8_64_add/;
@@ -237,11 +273,23 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x4_16_add sse2 neon dspr2/;
add_proto qw/void vp10_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x8_32_add/;
add_proto qw/void vp10_iht8x4_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x4_32_add/;
add_proto qw/void vp10_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht4x8_32_add/;
add_proto qw/void vp10_iht8x16_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x16_128_add/;
add_proto qw/void vp10_iht16x8_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht16x8_128_add/;
add_proto qw/void vp10_iht16x32_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht16x32_512_add/;
add_proto qw/void vp10_iht32x16_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht32x16_512_add/;
add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp10_iht8x8_64_add sse2 neon dspr2/;
@@ -382,11 +430,23 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vp10_highbd_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
specialize qw/vp10_highbd_iht4x4_16_add/;
add_proto qw/void vp10_highbd_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
specialize qw/vp10_highbd_iht4x8_32_add/;
add_proto qw/void vp10_highbd_iht8x4_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
specialize qw/vp10_highbd_iht8x4_32_add/;
add_proto qw/void vp10_highbd_iht4x8_32_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
specialize qw/vp10_highbd_iht4x8_32_add/;
add_proto qw/void vp10_highbd_iht8x16_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
specialize qw/vp10_highbd_iht8x16_128_add/;
add_proto qw/void vp10_highbd_iht16x8_128_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
specialize qw/vp10_highbd_iht16x8_128_add/;
add_proto qw/void vp10_highbd_iht16x32_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
specialize qw/vp10_highbd_iht16x32_512_add/;
add_proto qw/void vp10_highbd_iht32x16_512_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
specialize qw/vp10_highbd_iht32x16_512_add/;
add_proto qw/void vp10_highbd_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
specialize qw/vp10_highbd_iht8x8_64_add/;
@@ -439,11 +499,23 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vp10_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht4x4 sse2/;
add_proto qw/void vp10_fht4x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht4x8/;
add_proto qw/void vp10_fht8x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht8x4/;
add_proto qw/void vp10_fht4x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht4x8/;
add_proto qw/void vp10_fht8x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht8x16/;
add_proto qw/void vp10_fht16x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht16x8/;
add_proto qw/void vp10_fht16x32/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht16x32/;
add_proto qw/void vp10_fht32x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht32x16/;
add_proto qw/void vp10_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht8x8 sse2/;
@@ -460,11 +532,23 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vp10_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht4x4 sse2/;
add_proto qw/void vp10_fht4x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht4x8/;
add_proto qw/void vp10_fht8x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht8x4/;
add_proto qw/void vp10_fht4x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht4x8/;
add_proto qw/void vp10_fht8x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht8x16/;
add_proto qw/void vp10_fht16x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht16x8/;
add_proto qw/void vp10_fht16x32/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht16x32/;
add_proto qw/void vp10_fht32x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht32x16/;
add_proto qw/void vp10_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_fht8x8 sse2/;
@@ -743,11 +827,23 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vp10_highbd_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_highbd_fht4x4 sse4_1/;
add_proto qw/void vp10_highbd_fht4x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_highbd_fht4x8/;
add_proto qw/void vp10_highbd_fht8x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_highbd_fht8x4/;
add_proto qw/void vp10_highbd_fht4x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_highbd_fht4x8/;
add_proto qw/void vp10_highbd_fht8x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_highbd_fht8x16/;
add_proto qw/void vp10_highbd_fht16x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_highbd_fht16x8/;
add_proto qw/void vp10_highbd_fht16x32/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_highbd_fht16x32/;
add_proto qw/void vp10_highbd_fht32x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_highbd_fht32x16/;
add_proto qw/void vp10_highbd_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp10_highbd_fht8x8/;

249
vp10/encoder/dct.c
View File

@@ -1251,6 +1251,82 @@ static const transform_2d FHT_8x4[] = {
{ fadst4, fidtx8 }, // V_FLIPADST
{ fidtx4, fadst8 }, // H_FLIPADST
};
static const transform_2d FHT_8x16[] = {
{ fdct16, fdct8 }, // DCT_DCT
{ fadst16, fdct8 }, // ADST_DCT
{ fdct16, fadst8 }, // DCT_ADST
{ fadst16, fadst8 }, // ADST_ADST
{ fadst16, fdct8 }, // FLIPADST_DCT
{ fdct16, fadst8 }, // DCT_FLIPADST
{ fadst16, fadst8 }, // FLIPADST_FLIPADST
{ fadst16, fadst8 }, // ADST_FLIPADST
{ fadst16, fadst8 }, // FLIPADST_ADST
{ fidtx16, fidtx8 }, // IDTX
{ fdct16, fidtx8 }, // V_DCT
{ fidtx16, fdct8 }, // H_DCT
{ fadst16, fidtx8 }, // V_ADST
{ fidtx16, fadst8 }, // H_ADST
{ fadst16, fidtx8 }, // V_FLIPADST
{ fidtx16, fadst8 }, // H_FLIPADST
};
static const transform_2d FHT_16x8[] = {
{ fdct8, fdct16 }, // DCT_DCT
{ fadst8, fdct16 }, // ADST_DCT
{ fdct8, fadst16 }, // DCT_ADST
{ fadst8, fadst16 }, // ADST_ADST
{ fadst8, fdct16 }, // FLIPADST_DCT
{ fdct8, fadst16 }, // DCT_FLIPADST
{ fadst8, fadst16 }, // FLIPADST_FLIPADST
{ fadst8, fadst16 }, // ADST_FLIPADST
{ fadst8, fadst16 }, // FLIPADST_ADST
{ fidtx8, fidtx16 }, // IDTX
{ fdct8, fidtx16 }, // V_DCT
{ fidtx8, fdct16 }, // H_DCT
{ fadst8, fidtx16 }, // V_ADST
{ fidtx8, fadst16 }, // H_ADST
{ fadst8, fidtx16 }, // V_FLIPADST
{ fidtx8, fadst16 }, // H_FLIPADST
};
static const transform_2d FHT_16x32[] = {
{ fdct32, fdct16 }, // DCT_DCT
{ fhalfright32, fdct16 }, // ADST_DCT
{ fdct32, fadst16 }, // DCT_ADST
{ fhalfright32, fadst16 }, // ADST_ADST
{ fhalfright32, fdct16 }, // FLIPADST_DCT
{ fdct32, fadst16 }, // DCT_FLIPADST
{ fhalfright32, fadst16 }, // FLIPADST_FLIPADST
{ fhalfright32, fadst16 }, // ADST_FLIPADST
{ fhalfright32, fadst16 }, // FLIPADST_ADST
{ fidtx32, fidtx16 }, // IDTX
{ fdct32, fidtx16 }, // V_DCT
{ fidtx32, fdct16 }, // H_DCT
{ fhalfright32, fidtx16 }, // V_ADST
{ fidtx32, fadst16 }, // H_ADST
{ fhalfright32, fidtx16 }, // V_FLIPADST
{ fidtx32, fadst16 }, // H_FLIPADST
};
static const transform_2d FHT_32x16[] = {
{ fdct16, fdct32 }, // DCT_DCT
{ fadst16, fdct32 }, // ADST_DCT
{ fdct16, fhalfright32 }, // DCT_ADST
{ fadst16, fhalfright32 }, // ADST_ADST
{ fadst16, fdct32 }, // FLIPADST_DCT
{ fdct16, fhalfright32 }, // DCT_FLIPADST
{ fadst16, fhalfright32 }, // FLIPADST_FLIPADST
{ fadst16, fhalfright32 }, // ADST_FLIPADST
{ fadst16, fhalfright32 }, // FLIPADST_ADST
{ fidtx16, fidtx32 }, // IDTX
{ fdct16, fidtx32 }, // V_DCT
{ fidtx16, fdct32 }, // H_DCT
{ fadst16, fidtx32 }, // V_ADST
{ fidtx16, fhalfright32 }, // H_ADST
{ fadst16, fidtx32 }, // V_FLIPADST
{ fidtx16, fhalfright32 }, // H_FLIPADST
};
#endif // CONFIG_EXT_TX
void vp10_fht4x4_c(const int16_t *input, tran_low_t *output, int stride,
@@ -1299,10 +1375,12 @@ void vp10_fht4x8_c(const int16_t *input, tran_low_t *output, int stride,
// Columns
for (i = 0; i < n; ++i) {
for (j = 0; j < n2; ++j) temp_in[j] = input[j * stride + i] * 8;
for (j = 0; j < n2; ++j)
temp_in[j] = (tran_low_t)fdct_round_shift(
input[j * stride + i] * 8 * Sqrt2);
ht.cols(temp_in, temp_out);
for (j = 0; j < n2; ++j)
out[j * n + i] = (tran_low_t)fdct_round_shift(temp_out[j] * Sqrt2);
out[j * n + i] = temp_out[j];
}
// Rows
@@ -1327,10 +1405,12 @@ void vp10_fht8x4_c(const int16_t *input, tran_low_t *output, int stride,
// Columns
for (i = 0; i < n2; ++i) {
for (j = 0; j < n; ++j) temp_in[j] = input[j * stride + i] * 8;
for (j = 0; j < n; ++j)
temp_in[j] = (tran_low_t)fdct_round_shift(
input[j * stride + i] * 8 * Sqrt2);
ht.cols(temp_in, temp_out);
for (j = 0; j < n; ++j)
out[j * n2 + i] = (tran_low_t)fdct_round_shift(temp_out[j] * Sqrt2);
out[j * n2 + i] = temp_out[j];
}
// Rows
@@ -1341,6 +1421,137 @@ void vp10_fht8x4_c(const int16_t *input, tran_low_t *output, int stride,
}
// Note: overall scale factor of transform is 8 times unitary
}
void vp10_fht8x16_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
const int n = 8;
const int n2 = 16;
tran_low_t out[16 * 8];
tran_low_t temp_in[16], temp_out[16];
int i, j;
const transform_2d ht = FHT_8x16[tx_type];
int16_t flipped_input[16 * 8];
maybe_flip_input(&input, &stride, n2, n, flipped_input, tx_type);
// Columns
for (i = 0; i < n; ++i) {
for (j = 0; j < n2; ++j)
temp_in[j] = (tran_low_t)fdct_round_shift(
input[j * stride + i] * 4 * Sqrt2);
ht.cols(temp_in, temp_out);
for (j = 0; j < n2; ++j)
out[j * n + i] = temp_out[j];
}
// Rows
for (i = 0; i < n2; ++i) {
for (j = 0; j < n; ++j)
temp_in[j] = out[j + i * n];
ht.rows(temp_in, temp_out);
for (j = 0; j < n; ++j)
output[j + i * n] = (temp_out[j] + 1) >> 1;
}
// Note: overall scale factor of transform is 8 times unitary
}
void vp10_fht16x8_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
const int n = 8;
const int n2 = 16;
tran_low_t out[16 * 8];
tran_low_t temp_in[16], temp_out[16];
int i, j;
const transform_2d ht = FHT_16x8[tx_type];
int16_t flipped_input[16 * 8];
maybe_flip_input(&input, &stride, n, n2, flipped_input, tx_type);
// Columns
for (i = 0; i < n2; ++i) {
for (j = 0; j < n; ++j)
temp_in[j] = (tran_low_t)fdct_round_shift(
input[j * stride + i] * 4 * Sqrt2);
ht.cols(temp_in, temp_out);
for (j = 0; j < n; ++j)
out[j * n2 + i] = temp_out[j];
}
// Rows
for (i = 0; i < n; ++i) {
for (j = 0; j < n2; ++j)
temp_in[j] = out[j + i * n2];
ht.rows(temp_in, temp_out);
for (j = 0; j < n2; ++j)
output[j + i * n2] = (temp_out[j] + 1) >> 1;
}
// Note: overall scale factor of transform is 8 times unitary
}
void vp10_fht16x32_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
const int n = 16;
const int n2 = 32;
tran_low_t out[32 * 16];
tran_low_t temp_in[32], temp_out[32];
int i, j;
const transform_2d ht = FHT_16x32[tx_type];
int16_t flipped_input[32 * 16];
maybe_flip_input(&input, &stride, n2, n, flipped_input, tx_type);
// Columns
for (i = 0; i < n; ++i) {
for (j = 0; j < n2; ++j)
temp_in[j] = (tran_low_t)fdct_round_shift(
input[j * stride + i] * Sqrt2);
ht.cols(temp_in, temp_out);
for (j = 0; j < n2; ++j)
out[j * n + i] = temp_out[j];
}
// Rows
for (i = 0; i < n2; ++i) {
for (j = 0; j < n; ++j)
temp_in[j] = out[j + i * n];
ht.rows(temp_in, temp_out);
for (j = 0; j < n; ++j)
output[j + i * n] =
(tran_low_t)((temp_out[j] + 1 + (temp_out[j] < 0)) >> 2);
}
// Note: overall scale factor of transform is 4 times unitary
}
void vp10_fht32x16_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
const int n = 16;
const int n2 = 32;
tran_low_t out[32 * 16];
tran_low_t temp_in[32], temp_out[32];
int i, j;
const transform_2d ht = FHT_32x16[tx_type];
int16_t flipped_input[32 * 16];
maybe_flip_input(&input, &stride, n, n2, flipped_input, tx_type);
// Columns
for (i = 0; i < n2; ++i) {
for (j = 0; j < n; ++j)
temp_in[j] = (tran_low_t)fdct_round_shift(
input[j * stride + i] * Sqrt2);
ht.cols(temp_in, temp_out);
for (j = 0; j < n; ++j)
out[j * n2 + i] = temp_out[j];
}
// Rows
for (i = 0; i < n; ++i) {
for (j = 0; j < n2; ++j)
temp_in[j] = out[j + i * n2];
ht.rows(temp_in, temp_out);
for (j = 0; j < n2; ++j)
output[j + i * n2] =
(tran_low_t)((temp_out[j] + 1 + (temp_out[j] < 0)) >> 2);
}
// Note: overall scale factor of transform is 4 times unitary
}
#endif // CONFIG_EXT_TX
void vp10_fdct8x8_quant_c(const int16_t *input, int stride,
@@ -1578,14 +1789,34 @@ void vp10_highbd_fht4x4_c(const int16_t *input, tran_low_t *output, int stride,
}
#if CONFIG_EXT_TX
void vp10_highbd_fht8x4_c(const int16_t *input, tran_low_t *output, int stride,
int tx_type) {
void vp10_highbd_fht4x8_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
vp10_fht4x8_c(input, output, stride, tx_type);
}
void vp10_highbd_fht8x4_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
vp10_fht8x4_c(input, output, stride, tx_type);
}
void vp10_highbd_fht4x8_c(const int16_t *input, tran_low_t *output, int stride,
int tx_type) {
vp10_fht4x8_c(input, output, stride, tx_type);
void vp10_highbd_fht8x16_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
vp10_fht8x16_c(input, output, stride, tx_type);
}
void vp10_highbd_fht16x8_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
vp10_fht16x8_c(input, output, stride, tx_type);
}
void vp10_highbd_fht16x32_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
vp10_fht16x32_c(input, output, stride, tx_type);
}
void vp10_highbd_fht32x16_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
vp10_fht32x16_c(input, output, stride, tx_type);
}
#endif // CONFIG_EXT_TX

131
vp10/encoder/hybrid_fwd_txfm.c
View File

@@ -55,19 +55,47 @@ static void fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
}
#if CONFIG_EXT_TX
static void fwd_txfm_8x4(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt) {
(void)fwd_txfm_opt;
vp10_fht8x4(src_diff, coeff, diff_stride, tx_type);
}
static void fwd_txfm_4x8(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt) {
(void)fwd_txfm_opt;
(void) fwd_txfm_opt;
vp10_fht4x8(src_diff, coeff, diff_stride, tx_type);
}
static void fwd_txfm_8x4(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt) {
(void) fwd_txfm_opt;
vp10_fht8x4(src_diff, coeff, diff_stride, tx_type);
}
static void fwd_txfm_8x16(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt) {
(void) fwd_txfm_opt;
vp10_fht8x16(src_diff, coeff, diff_stride, tx_type);
}
static void fwd_txfm_16x8(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt) {
(void) fwd_txfm_opt;
vp10_fht16x8(src_diff, coeff, diff_stride, tx_type);
}
static void fwd_txfm_16x32(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt) {
(void) fwd_txfm_opt;
vp10_fht16x32(src_diff, coeff, diff_stride, tx_type);
}
static void fwd_txfm_32x16(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt) {
(void) fwd_txfm_opt;
vp10_fht32x16(src_diff, coeff, diff_stride, tx_type);
}
#endif // CONFIG_EXT_TX
static void fwd_txfm_8x8(const int16_t *src_diff, tran_low_t *coeff,
@@ -213,21 +241,53 @@ static void highbd_fwd_txfm_4x4(const int16_t *src_diff, tran_low_t *coeff,
}
#if CONFIG_EXT_TX
static void highbd_fwd_txfm_8x4(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt, const int bd) {
(void)fwd_txfm_opt;
(void)bd;
vp10_highbd_fht8x4(src_diff, coeff, diff_stride, tx_type);
}
static void highbd_fwd_txfm_4x8(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt, const int bd) {
(void)fwd_txfm_opt;
(void)bd;
(void) fwd_txfm_opt;
(void) bd;
vp10_highbd_fht4x8(src_diff, coeff, diff_stride, tx_type);
}
static void highbd_fwd_txfm_8x4(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt, const int bd) {
(void) fwd_txfm_opt;
(void) bd;
vp10_highbd_fht8x4(src_diff, coeff, diff_stride, tx_type);
}
static void highbd_fwd_txfm_8x16(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt, const int bd) {
(void) fwd_txfm_opt;
(void) bd;
vp10_highbd_fht8x16(src_diff, coeff, diff_stride, tx_type);
}
static void highbd_fwd_txfm_16x8(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt, const int bd) {
(void) fwd_txfm_opt;
(void) bd;
vp10_highbd_fht16x8(src_diff, coeff, diff_stride, tx_type);
}
static void highbd_fwd_txfm_16x32(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt, const int bd) {
(void) fwd_txfm_opt;
(void) bd;
vp10_highbd_fht16x32(src_diff, coeff, diff_stride, tx_type);
}
static void highbd_fwd_txfm_32x16(const int16_t *src_diff, tran_low_t *coeff,
int diff_stride, TX_TYPE tx_type,
FWD_TXFM_OPT fwd_txfm_opt, const int bd) {
(void) fwd_txfm_opt;
(void) bd;
vp10_highbd_fht32x16(src_diff, coeff, diff_stride, tx_type);
}
#endif // CONFIG_EXT_TX
static void highbd_fwd_txfm_8x8(const int16_t *src_diff, tran_low_t *coeff,
@@ -361,6 +421,18 @@ void fwd_txfm(const int16_t *src_diff, tran_low_t *coeff, int diff_stride,
case TX_8X4:
fwd_txfm_8x4(src_diff, coeff, diff_stride, tx_type, fwd_txfm_opt);
break;
case TX_8X16:
fwd_txfm_8x16(src_diff, coeff, diff_stride, tx_type, fwd_txfm_opt);
break;
case TX_16X8:
fwd_txfm_16x8(src_diff, coeff, diff_stride, tx_type, fwd_txfm_opt);
break;
case TX_16X32:
fwd_txfm_16x32(src_diff, coeff, diff_stride, tx_type, fwd_txfm_opt);
break;
case TX_32X16:
fwd_txfm_32x16(src_diff, coeff, diff_stride, tx_type, fwd_txfm_opt);
break;
#endif // CONFIG_EXT_TX
case TX_4X4:
fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type, lossless);
@@ -400,11 +472,30 @@ void highbd_fwd_txfm(const int16_t *src_diff, tran_low_t *coeff,
highbd_fwd_txfm_8x4(src_diff, coeff, diff_stride, tx_type, fwd_txfm_opt,
bd);
break;
case TX_8X16:
highbd_fwd_txfm_8x16(src_diff, coeff, diff_stride, tx_type,
fwd_txfm_opt, bd);
break;
case TX_16X8:
highbd_fwd_txfm_16x8(src_diff, coeff, diff_stride, tx_type,
fwd_txfm_opt, bd);
break;
case TX_16X32:
highbd_fwd_txfm_16x32(src_diff, coeff, diff_stride, tx_type,
fwd_txfm_opt, bd);
break;
case TX_32X16:
highbd_fwd_txfm_32x16(src_diff, coeff, diff_stride, tx_type,
fwd_txfm_opt, bd);
break;
#endif // CONFIG_EXT_TX
case TX_4X4:
highbd_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type, lossless, bd);
highbd_fwd_txfm_4x4(src_diff, coeff, diff_stride, tx_type,
lossless, bd);
break;
default:
assert(0);
break;
default: assert(0); break;
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH

52
vp10/encoder/rd.c
View File

@@ -571,18 +571,6 @@ static void get_entropy_contexts_plane(
memcpy(t_above, above, sizeof(ENTROPY_CONTEXT) * num_4x4_w);
memcpy(t_left, left, sizeof(ENTROPY_CONTEXT) * num_4x4_h);
break;
#if CONFIG_EXT_TX
case TX_4X8:
memcpy(t_above, above, sizeof(ENTROPY_CONTEXT) * num_4x4_w);
for (i = 0; i < num_4x4_h; i += 2)
t_left[i] = !!*(const uint16_t *)&left[i];
break;
case TX_8X4:
for (i = 0; i < num_4x4_w; i += 2)
t_above[i] = !!*(const uint16_t *)&above[i];
memcpy(t_left, left, sizeof(ENTROPY_CONTEXT) * num_4x4_h);
break;
#endif // CONFIG_EXT_TX
case TX_8X8:
for (i = 0; i < num_4x4_w; i += 2)
t_above[i] = !!*(const uint16_t *)&above[i];
@@ -601,7 +589,45 @@ static void get_entropy_contexts_plane(
for (i = 0; i < num_4x4_h; i += 8)
t_left[i] = !!*(const uint64_t *)&left[i];
break;
default: assert(0 && "Invalid transform size."); break;
#if CONFIG_EXT_TX
case TX_4X8:
memcpy(t_above, above, sizeof(ENTROPY_CONTEXT) * num_4x4_w);
for (i = 0; i < num_4x4_h; i += 2)
t_left[i] = !!*(const uint16_t *)&left[i];
break;
case TX_8X4:
for (i = 0; i < num_4x4_w; i += 2)
t_above[i] = !!*(const uint16_t *)&above[i];
memcpy(t_left, left, sizeof(ENTROPY_CONTEXT) * num_4x4_h);
break;
case TX_8X16:
for (i = 0; i < num_4x4_w; i += 2)
t_above[i] = !!*(const uint16_t *)&above[i];
for (i = 0; i < num_4x4_h; i += 4)
t_left[i] = !!*(const uint32_t *)&left[i];
break;
case TX_16X8:
for (i = 0; i < num_4x4_w; i += 4)
t_above[i] = !!*(const uint32_t *)&above[i];
for (i = 0; i < num_4x4_h; i += 2)
t_left[i] = !!*(const uint16_t *)&left[i];
break;
case TX_16X32:
for (i = 0; i < num_4x4_w; i += 4)
t_above[i] = !!*(const uint32_t *)&above[i];
for (i = 0; i < num_4x4_h; i += 8)
t_left[i] = !!*(const uint64_t *)&left[i];
break;
case TX_32X16:
for (i = 0; i < num_4x4_w; i += 8)
t_above[i] = !!*(const uint64_t *)&above[i];
for (i = 0; i < num_4x4_h; i += 4)
t_left[i] = !!*(const uint32_t *)&left[i];
break;
#endif // CONFIG_EXT_TX
default:
assert(0 && "Invalid transform size.");
break;
}
}

16
vp10/encoder/rdopt.c
View File

@@ -1107,6 +1107,22 @@ static uint64_t sum_squares_2d(const int16_t *diff, int diff_stride,
sse = vpx_sum_squares_2d_i16(diff, diff_stride, 4) +
vpx_sum_squares_2d_i16(diff + 4, diff_stride, 4);
break;
case TX_8X16:
sse = vpx_sum_squares_2d_i16(diff, diff_stride, 8) +
vpx_sum_squares_2d_i16(diff + 8 * diff_stride, diff_stride, 8);
break;
case TX_16X8:
sse = vpx_sum_squares_2d_i16(diff, diff_stride, 8) +
vpx_sum_squares_2d_i16(diff + 8, diff_stride, 8);
break;
case TX_16X32:
sse = vpx_sum_squares_2d_i16(diff, diff_stride, 16) +
vpx_sum_squares_2d_i16(diff + 16 * diff_stride, diff_stride, 16);
break;
case TX_32X16:
sse = vpx_sum_squares_2d_i16(diff, diff_stride, 16) +
vpx_sum_squares_2d_i16(diff + 16, diff_stride, 16);
break;
#endif // CONFIG_EXT_TX
default:
assert(tx_size < TX_SIZES);