Moved from nextgenv2 branch to test with other experiments.
derflr: +1.629%
Change-Id: Ie7c720053ed8b628177679c4351bb31b54716a71
static TX_TYPE ext_tx_to_txtype[EXT_TX_TYPES] = {
DCT_DCT,
- ADST_ADST,
- FLIPADST_FLIPADST,
- ADST_FLIPADST,
- FLIPADST_ADST,
ADST_DCT,
DCT_ADST,
+ ADST_ADST,
FLIPADST_DCT,
DCT_FLIPADST,
+ FLIPADST_FLIPADST,
+ ADST_FLIPADST,
+ FLIPADST_ADST,
#if CONFIG_DST1
DST_DST,
-#endif
+ DST_DCT,
+ DCT_DST,
+ DST_ADST,
+ ADST_DST,
+ DST_FLIPADST,
+ FLIPADST_DST,
+#endif // CONFIG_DST1
};
+#if CONFIG_DST1
+static INLINE int is_dst_used(TX_TYPE tx_type) {
+ return (tx_type == DST_DST ||
+ tx_type == DST_DCT || tx_type == DCT_DST ||
+ tx_type == DST_ADST || tx_type == ADST_DST ||
+ tx_type == DST_FLIPADST || tx_type == FLIPADST_DST);
+}
+#endif // CONFIG_DST1
+
#if CONFIG_WAVELETS
static TX_TYPE ext_tx_to_txtype_large[EXT_TX_TYPES_LARGE] = {
DCT_DCT,
const vp9_tree_index vp9_ext_tx_tree[TREE_SIZE(EXT_TX_TYPES)] = {
-NORM, 2,
-ALT9, 4,
- 6, 12,
+ 6, 16,
8, 10,
+ -ALT10, -ALT11,
+ 12, 14,
-ALT1, -ALT2,
- -ALT3, -ALT4,
- 14, 16,
- -ALT5, -ALT6,
- -ALT7, -ALT8,
+ -ALT4, -ALT5,
+ 18, 24,
+ 20, 22,
+ -ALT12, -ALT13,
+ -ALT14, -ALT15,
+ 26, 28,
+ -ALT3, -ALT6,
+ -ALT7, -ALT8
};
#if CONFIG_WAVELETS
static const vp9_prob default_ext_tx_prob[TX_SIZES][EXT_TX_TYPES - 1] = {
- { 224, 28, 128, 128, 128, 128, 128, 128, 128 },
- { 200, 28, 128, 128, 128, 128, 128, 128, 128 },
- { 176, 28, 128, 128, 128, 128, 128, 128, 128 },
- { 160, 28, 128, 128, 128, 128, 128, 128, 128 },
+ { 216, 20, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 },
+ { 192, 20, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 },
+ { 168, 20, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 },
+ { 160, 20, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 },
#if CONFIG_TX64X64
- { 160, 28, 128, 128, 128, 128, 128, 128, 128 },
+ { 160, 20, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 },
#endif // CONFIG_TX64X64
};
#else
static const vp9_prob default_ext_tx_prob[3][EXT_TX_TYPES - 1] = {
- { 224, 28, 128, 128, 128, 128, 128, 128, 128 },
- { 200, 28, 128, 128, 128, 128, 128, 128, 128 },
- { 176, 28, 128, 128, 128, 128, 128, 128, 128 },
+ { 144, 12, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 },
+ { 128, 12, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 },
+ { 112, 12, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 },
};
#endif // CONFIG_WAVELETS
{ 176, 128, 128, 128, 128, 128, 128, 128 },
};
#endif // CONFIG_WAVELETS
-
#endif // CONFIG_DST1
#if CONFIG_WAVELETS
ADST_FLIPADST = 7,
FLIPADST_ADST = 8,
#if CONFIG_DST1
- DST_DST,
+ DST_DST = 9,
+ DST_DCT = 10,
+ DCT_DST = 11,
+ DST_ADST = 12,
+ ADST_DST = 13,
+ DST_FLIPADST = 14,
+ FLIPADST_DST = 15,
#endif // CONFIG_DST1
#if CONFIG_WAVELETS
WAVELET1_DCT_DCT,
ALT8 = 8,
#if CONFIG_DST1
ALT9 = 9,
+ ALT10 = 10,
+ ALT11 = 11,
+ ALT12 = 12,
+ ALT13 = 13,
+ ALT14 = 14,
+ ALT15 = 15,
#endif // CONFIG_DST1
EXT_TX_TYPES
} EXT_TX_TYPE;
#include "vp9/common/vp9_blockd.h"
#include "vp9/common/vp9_idct.h"
+#if CONFIG_EXT_TX
#if CONFIG_DST1
-// Integers to represent double.
-// The sine transform formula is: X{i} = Sum_{0<=j<N}( x_j *
-// sin((i+1)*(j+1)/(N+1) * PI) ) * sqrt(2/(N+1))
-// e.g. when N == 4, it is a series of sin(PI*i/5) and sqrt(2/5). Similar for
-// N = 8 and 16.
-// For integer calculation, we multiply 2^14.
-// sin_pi_5 = sin(PI/5)*pow(2,14)
-// sqrt_2_5 = sqrt(2/5)*pow(2,14)
-
-// {sin(pi/5), sin(pi*2/5)}
-int sinvalue_lookup_table_4[2] = { 9630, 15582 };
-// {sin(pi/9), sin(pi*2/9), ..., sin(pi*4/9)}
-int sinvalue_lookup_table_8[4] = { 5604, 10531, 14189, 16135 };
-// {sin(pi/17), ...
-int sinvalue_lookup_table_16[] = { 3011, 5919, 8625, 11038,
- 13075, 14666, 15759, 16314 };
-
-void vp9_dst1d_type1(int64_t *in, int64_t *out, int N) {
+static INLINE void idst_core(const tran_low_t *input, tran_low_t *output,
+ const int N, const int32_t *dst_lookup,
+ int bd) {
int i, j;
+ (void) bd;
for (i = 0; i < N; i++) {
int64_t sum = 0;
for (j = 0; j < N; j++) {
- int64_t sinvalue = 0;
int idx = (i + 1) * (j + 1);
int sign = 0;
if (idx > N + 1) {
- sign = idx / (N + 1);
- sign = sign % 2 ? 1 : 0;
+ sign = (idx / (N + 1)) & 1;
idx %= (N + 1);
}
- idx = idx > N + 1 - idx ? N + 1 - idx : idx;
+ idx = MIN(idx, N + 1 - idx);
if (idx == 0) continue;
- idx--;
-
- if (N == 4)
- sinvalue = sinvalue_lookup_table_4[idx];
- else if (N == 8)
- sinvalue = sinvalue_lookup_table_8[idx];
- else if (N == 16)
- sinvalue = sinvalue_lookup_table_16[idx];
- else
- assert(0 && "Invalid transform size.");
- if (sign) sinvalue = -sinvalue;
-
- sum += in[j] * sinvalue;
+ sum += (int64_t)input[j] * dst_lookup[idx - 1] * (sign ? -1 : 1);
}
- out[i] = sum;
+ sum = ROUND_POWER_OF_TWO(sum, (2 * DCT_CONST_BITS));
+ output[i] = WRAPLOW(sum, bd);
}
}
-static void idstNxN_add(const tran_low_t *input, uint8_t *output,
- int stride, int N) {
- const int val_2_5 = 6554;
- const int val_2_9 = 3641;
- const int val_2_17 = 1928;
- int i, j;
- int64_t *in = (int64_t *) malloc (N * sizeof(int64_t));
- int64_t *inter = (int64_t *) malloc (N * sizeof(int64_t));
- int64_t *mat = (int64_t *) malloc (N * N * sizeof(int64_t));
- int64_t *mat2 = (int64_t *) malloc (N * N * sizeof(int64_t));
- int64_t val;
-
- // 1d dst: transform columns
- for (j = 0; j < N; j++) {
- for (i = 0; i < N; i++) {
- in[i] = input[i * N + j];
- }
- vp9_dst1d_type1(in, inter, N);
- for (i = 0; i < N; i++) {
- mat2[i * N + j] = inter[i];
- }
- }
+void idst4(const tran_low_t *input, tran_low_t *output) {
+ idst_core(input, output, 4, dst_lookup4, 8);
+}
- // transpose
- for (i = 0; i < N; i++)
- for (j = 0; j < N; j++)
- mat[i * N + j] = mat2[i + j * N];
-
- switch (N) {
- case 4:
- val = val_2_5;
- break;
- case 8:
- val = val_2_9;
- break;
- case 16:
- val = val_2_17;
- break;
- default:
- assert(0 && "Invalid transform size.");
- return;
- }
+void idst8(const tran_low_t *input, tran_low_t *output) {
+ idst_core(input, output, 8, dst_lookup8, 8);
+}
- // 1d dst: transform rows
- for (j = 0; j < N; j++) {
- for (i = 0; i < N; i++) {
- in[i] = mat[i * N + j];
- }
- vp9_dst1d_type1(in, inter, N);
- for (i = 0; i < N; i++) {
- int64_t tmp = inter[i];
- tmp = tmp >> DCT_CONST_BITS;
- tmp *= val;
- mat[i*N + j] = tmp >> (2 * DCT_CONST_BITS);
- }
- }
- for (i = 0; i < N; i++) {
- for (j = 0; j < N; j++) {
- tran_high_t tmp = mat[i * N + j];
- tmp = WRAPLOW(tmp, 8);
- output[i * stride + j] = clip_pixel_add(output[i * stride + j],
- ROUND_POWER_OF_TWO(tmp, 3));
- }
- }
- free(in);
- free(inter);
- free(mat);
- free(mat2);
+void idst16(const tran_low_t *input, tran_low_t *output) {
+ idst_core(input, output, 16, dst_lookup16, 8);
}
-void vp9_idst4x4_add(const tran_low_t *input, uint8_t *dest, int stride) {
- idstNxN_add(input, dest, stride, 4);
+#if CONFIG_VP9_HIGHBITDEPTH
+void highbd_idst4(const tran_low_t *input, tran_low_t *output, int bd) {
+ idst_core(input, output, 4, dst_lookup4, bd);
}
-void vp9_idst8x8_add(const tran_low_t *input, uint8_t *dest, int stride) {
- idstNxN_add(input, dest, stride, 8);
+void highbd_idst8(const tran_low_t *input, tran_low_t *output, int bd) {
+ idst_core(input, output, 8, dst_lookup8, bd);
}
-void vp9_idst16x16_add(const tran_low_t *input, uint8_t *dest, int stride) {
- idstNxN_add(input, dest, stride, 16);
+void highbd_idst16(const tran_low_t *input, tran_low_t *output, int bd) {
+ idst_core(input, output, 16, dst_lookup16, bd);
}
+#endif // CONFIG_VP9_HIGHBITDEPTH
#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
void vp9_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
/* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
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, idct4 }, // DCT_DCT = 0
- { iadst4, idct4 }, // ADST_DCT = 1
- { idct4, iadst4 }, // DCT_ADST = 2
- { iadst4, iadst4 } // ADST_ADST = 3
+ { idct4, idct4 }, // DCT_DCT = 0
+ { iadst4, idct4 }, // ADST_DCT = 1
+ { idct4, iadst4 }, // DCT_ADST = 2
+ { iadst4, iadst4 }, // ADST_ADST = 3
+#if CONFIG_EXT_TX
+ { iadst4, idct4 }, // FLIPADST_DCT = 4
+ { idct4, iadst4 }, // DCT_FLIPADST = 5
+ { iadst4, iadst4 }, // FLIPADST_FLIPADST = 6
+ { iadst4, iadst4 }, // ADST_FLIPADST = 7
+ { iadst4, iadst4 }, // FLIPADST_ADST = 8
+#if CONFIG_DST1
+ { idst4, idst4 }, // DST_DST = 9
+ { idst4, idct4 }, // DST_DCT = 10
+ { idct4, idst4 }, // DCT_DST = 11
+ { idst4, iadst4 }, // DST_ADST = 12
+ { iadst4, idst4 }, // ADST_DST = 13
+ { idst4, iadst4 }, // DST_FLIPADST = 14
+ { iadst4, idst4 }, // FLIPADST_DST = 15
+#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
};
int i, j;
{ idct8, idct8 }, // DCT_DCT = 0
{ iadst8, idct8 }, // ADST_DCT = 1
{ idct8, iadst8 }, // DCT_ADST = 2
- { iadst8, iadst8 } // ADST_ADST = 3
+ { iadst8, iadst8 }, // ADST_ADST = 3
+#if CONFIG_EXT_TX
+ { iadst8, idct8 }, // FLIPADST_DCT = 4
+ { idct8, iadst8 }, // DCT_FLIPADST = 5
+ { iadst8, iadst8 }, // FLIPADST_FLIPADST = 6
+ { iadst8, iadst8 }, // ADST_FLIPADST = 7
+ { iadst8, iadst8 }, // FLIPADST_ADST = 8
+#if CONFIG_DST1
+ { idst8, idst8 }, // DST_DST = 9
+ { idst8, idct8 }, // DST_DCT = 10
+ { idct8, idst8 }, // DCT_DST = 11
+ { idst8, iadst8 }, // DST_ADST = 12
+ { iadst8, idst8 }, // ADST_DST = 13
+ { idst8, iadst8 }, // DST_FLIPADST = 14
+ { iadst8, idst8 }, // FLIPADST_DST = 15
+#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
};
#if CONFIG_EXT_TX
{ idct16, idct16 }, // DCT_DCT = 0
{ iadst16, idct16 }, // ADST_DCT = 1
{ idct16, iadst16 }, // DCT_ADST = 2
- { iadst16, iadst16 } // ADST_ADST = 3
+ { iadst16, iadst16 }, // ADST_ADST = 3
+#if CONFIG_EXT_TX
+ { iadst16, idct16 }, // FLIPADST_DCT = 4
+ { idct16, iadst16 }, // DCT_FLIPADST = 5
+ { iadst16, iadst16 }, // FLIPADST_FLIPADST = 6
+ { iadst16, iadst16 }, // ADST_FLIPADST = 7
+ { iadst16, iadst16 }, // FLIPADST_ADST = 8
+#if CONFIG_DST1
+ { idst16, idst16 }, // DST_DST = 9
+ { idst16, idct16 }, // DST_DCT = 10
+ { idct16, idst16 }, // DCT_DST = 11
+ { idst16, iadst16 }, // DST_ADST = 12
+ { iadst16, idst16 }, // ADST_DST = 13
+ { idst16, iadst16 }, // DST_FLIPADST = 14
+ { iadst16, idst16 }, // FLIPADST_DST = 15
+#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
};
void vp9_iht16x16_256_add_c(const tran_low_t *input, uint8_t *dest, int stride,
vp9_idct4x4_add(input, dest, stride, eob);
#if CONFIG_EXT_TX
#if CONFIG_DST1
- } else if (tx_type == DST_DST) {
- vp9_idst4x4_add(input, dest, stride);
+ } else if (is_dst_used(tx_type)) {
+ vp9_iht4x4_16_add_c(input, dest, stride, tx_type);
#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
flipud(dest, stride, 4);
vp9_idct8x8_add(input, dest, stride, eob);
#if CONFIG_EXT_TX
#if CONFIG_DST1
- } else if (tx_type == DST_DST) {
- vp9_idst8x8_add(input, dest, stride);
+ } else if (is_dst_used(tx_type)) {
+ vp9_iht8x8_64_add_c(input, dest, stride, tx_type);
#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
flipud(dest, stride, 8);
vp9_idct16x16_add(input, dest, stride, eob);
#if CONFIG_EXT_TX
#if CONFIG_DST1
- } else if (tx_type == DST_DST) {
- vp9_idst16x16_add(input, dest, stride);
+ } else if (is_dst_used(tx_type)) {
+ vp9_iht16x16_256_add_c(input, dest, stride, tx_type);
#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
flipud(dest, stride, 16);
{ vp9_highbd_idct4, vp9_highbd_idct4 }, // DCT_DCT = 0
{ highbd_iadst4, vp9_highbd_idct4 }, // ADST_DCT = 1
{ vp9_highbd_idct4, highbd_iadst4 }, // DCT_ADST = 2
- { highbd_iadst4, highbd_iadst4 } // ADST_ADST = 3
+ { highbd_iadst4, highbd_iadst4 }, // ADST_ADST = 3
+#if CONFIG_EXT_TX
+ { highbd_iadst4, vp9_highbd_idct4 }, // FLIPADST_DCT = 4
+ { vp9_highbd_idct4, highbd_iadst4 }, // DCT_FLIPADST = 5
+ { highbd_iadst4, highbd_iadst4 }, // FLIPADST_FLIPADST = 6
+ { highbd_iadst4, highbd_iadst4 }, // ADST_FLIPADST = 7
+ { highbd_iadst4, highbd_iadst4 }, // FLIPADST_ADST = 8
+#if CONFIG_DST1
+ { highbd_idst4, highbd_idst4 }, // DST_DST = 9
+ { highbd_idst4, vp9_highbd_idct4 }, // DST_DCT = 10
+ { vp9_highbd_idct4, highbd_idst4 }, // DCT_DST = 11
+ { highbd_idst4, highbd_iadst4 }, // DST_ADST = 12
+ { highbd_iadst4, highbd_idst4 }, // ADST_DST = 13
+ { highbd_idst4, highbd_iadst4 }, // DST_FLIPADST = 14
+ { highbd_iadst4, highbd_idst4 }, // FLIPADST_DST = 15
+#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
};
uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
{ vp9_highbd_idct8, vp9_highbd_idct8 }, // DCT_DCT = 0
{ highbd_iadst8, vp9_highbd_idct8 }, // ADST_DCT = 1
{ vp9_highbd_idct8, highbd_iadst8 }, // DCT_ADST = 2
- { highbd_iadst8, highbd_iadst8 } // ADST_ADST = 3
+ { highbd_iadst8, highbd_iadst8 }, // ADST_ADST = 3
+#if CONFIG_EXT_TX
+ { highbd_iadst8, vp9_highbd_idct8 }, // FLIPADST_DCT = 4
+ { vp9_highbd_idct8, highbd_iadst8 }, // DCT_FLIPADST = 5
+ { highbd_iadst8, highbd_iadst8 }, // FLIPADST_FLIPADST = 6
+ { highbd_iadst8, highbd_iadst8 }, // ADST_FLIPADST = 7
+ { highbd_iadst8, highbd_iadst8 }, // FLIPADST_ADST = 8
+#if CONFIG_DST1
+ { highbd_idst8, highbd_idst8 }, // DST_DST = 9
+ { highbd_idst8, vp9_highbd_idct8 }, // DST_DCT = 10
+ { vp9_highbd_idct8, highbd_idst8 }, // DCT_DST = 11
+ { highbd_idst8, highbd_iadst8 }, // DST_ADST = 12
+ { highbd_iadst8, highbd_idst8 }, // ADST_DST = 13
+ { highbd_idst8, highbd_iadst8 }, // DST_FLIPADST = 14
+ { highbd_iadst8, highbd_idst8 }, // FLIPADST_DST = 15
+#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
};
void vp9_highbd_iht8x8_64_add_c(const tran_low_t *input, uint8_t *dest8,
{ vp9_highbd_idct16, vp9_highbd_idct16 }, // DCT_DCT = 0
{ highbd_iadst16, vp9_highbd_idct16 }, // ADST_DCT = 1
{ vp9_highbd_idct16, highbd_iadst16 }, // DCT_ADST = 2
- { highbd_iadst16, highbd_iadst16 } // ADST_ADST = 3
+ { highbd_iadst16, highbd_iadst16 }, // ADST_ADST = 3
+#if CONFIG_EXT_TX
+ { highbd_iadst16, vp9_highbd_idct16 }, // FLIPADST_DCT = 4
+ { vp9_highbd_idct16, highbd_iadst16 }, // DCT_FLIPADST = 5
+ { highbd_iadst16, highbd_iadst16 }, // FLIPADST_FLIPADST = 6
+ { highbd_iadst16, highbd_iadst16 }, // ADST_FLIPADST = 7
+ { highbd_iadst16, highbd_iadst16 }, // FLIPADST_ADST = 8
+#if CONFIG_DST1
+ { highbd_idst16, highbd_idst16 }, // DST_DST = 9
+ { highbd_idst16, vp9_highbd_idct16 }, // DST_DCT = 10
+ { vp9_highbd_idct16, highbd_idst16 }, // DCT_DST = 11
+ { highbd_idst16, highbd_iadst16 }, // DST_ADST = 12
+ { highbd_iadst16, highbd_idst16 }, // ADST_DST = 13
+ { highbd_idst16, highbd_iadst16 }, // DST_FLIPADST = 14
+ { highbd_iadst16, highbd_idst16 }, // FLIPADST_DST = 15
+#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
};
void vp9_highbd_iht16x16_256_add_c(const tran_low_t *input, uint8_t *dest8,
if (tx_type == DCT_DCT) {
vp9_highbd_idct4x4_add(input, dest, stride, eob, bd);
#if CONFIG_EXT_TX
+#if CONFIG_DST1
+ } else if (is_dst_used(tx_type)) {
+ vp9_highbd_iht4x4_16_add_c(input, dest, stride, tx_type, bd);
+#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
flipud16(CONVERT_TO_SHORTPTR(dest), stride, 4);
vp9_highbd_iht4x4_16_add(input, dest, stride, ADST_DCT, bd);
if (tx_type == DCT_DCT) {
vp9_highbd_idct8x8_add(input, dest, stride, eob, bd);
#if CONFIG_EXT_TX
+#if CONFIG_DST1
+ } else if (is_dst_used(tx_type)) {
+ vp9_highbd_iht8x8_64_add_c(input, dest, stride, tx_type, bd);
+#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
flipud16(CONVERT_TO_SHORTPTR(dest), stride, 8);
vp9_highbd_iht8x8_64_add(input, dest, stride, ADST_DCT, bd);
if (tx_type == DCT_DCT) {
vp9_highbd_idct16x16_add(input, dest, stride, eob, bd);
#if CONFIG_EXT_TX
+#if CONFIG_DST1
+ } else if (is_dst_used(tx_type)) {
+ vp9_highbd_iht16x16_256_add_c(input, dest, stride, tx_type, bd);
+#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
flipud16(CONVERT_TO_SHORTPTR(dest), stride, 16);
vp9_highbd_iht16x16_256_add(input, dest, stride, ADST_DCT, bd);
void vp9_iht4x4_16_c(const tran_low_t *input, int16_t *dest, int stride,
int tx_type) {
const transform_2d IHT_4[] = {
- { idct4, idct4 }, // DCT_DCT = 0
- { iadst4, idct4 }, // ADST_DCT = 1
- { idct4, iadst4 }, // DCT_ADST = 2
- { iadst4, iadst4 } // ADST_ADST = 3
+ { idct4, idct4 }, // DCT_DCT = 0
+ { iadst4, idct4 }, // ADST_DCT = 1
+ { idct4, iadst4 }, // DCT_ADST = 2
+ { iadst4, iadst4 }, // ADST_ADST = 3
+#if CONFIG_EXT_TX
+ { iadst4, idct4 }, // FLIPADST_DCT = 4
+ { idct4, iadst4 }, // DCT_FLIPADST = 5
+ { iadst4, iadst4 }, // FLIPADST_FLIPADST = 6
+ { iadst4, iadst4 }, // ADST_FLIPADST = 7
+ { iadst4, iadst4 }, // FLIPADST_ADST = 8
+#if CONFIG_DST1
+ { idst4, idst4 }, // DST_DST = 9
+ { idst4, idct4 }, // DST_DCT = 10
+ { idct4, idst4 }, // DCT_DST = 11
+ { idst4, iadst4 }, // DST_ADST = 12
+ { iadst4, idst4 }, // ADST_DST = 13
+ { idst4, iadst4 }, // DST_FLIPADST = 14
+ { iadst4, idst4 }, // FLIPADST_DST = 15
+#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
};
int i, j;
if (tx_type == DCT_DCT) {
vp9_idct4x4(input, dest, stride, eob);
#if CONFIG_EXT_TX
+#if CONFIG_DST1
+ } else if (is_dst_used(tx_type)) {
+ vp9_iht4x4_16_c(input, dest, stride, tx_type);
+#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
flipud(dest, stride, 4);
vp9_iht4x4_16(input, dest, stride, ADST_DCT);
if (tx_type == DCT_DCT) {
vp9_idct8x8(input, dest, stride, eob);
#if CONFIG_EXT_TX
+#if CONFIG_DST1
+ } else if (is_dst_used(tx_type)) {
+ vp9_iht8x8_64_c(input, dest, stride, tx_type);
+#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
flipud(dest, stride, 8);
vp9_iht8x8_64(input, dest, stride, ADST_DCT);
if (tx_type == DCT_DCT) {
vp9_idct16x16(input, dest, stride, eob);
#if CONFIG_EXT_TX
+#if CONFIG_DST1
+ } else if (is_dst_used(tx_type)) {
+ vp9_iht16x16_256_c(input, dest, stride, tx_type);
+#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
flipud(dest, stride, 16);
vp9_iht16x16_256(input, dest, stride, ADST_DCT);
static const tran_high_t sinpi_3_9 = 13377;
static const tran_high_t sinpi_4_9 = 15212;
+#if CONFIG_EXT_TX
+#if CONFIG_DST1
+static const int32_t dst_lookup4[] = {
+ // {sin(pi/5), sin(pi*2/5)} * sqrt(2/5) * sqrt(2)
+ // at precision of 2 * DCT_CONST_BITS bits
+ 141124871, 228344838,
+};
+
+static const int32_t dst_lookup8[] = {
+ // {sin(pi/9), sin(pi*2/9), ..., sin(pi*4/9)} * sqrt(2/9) * 2
+ // at precision of 2 * DCT_CONST_BITS bits
+ 86559612, 162678858, 219176632, 249238470
+};
+
+static const int32_t dst_lookup16[] = {
+ // {sin(pi/17), sin(pi*2/17, ..., sin(pi*8/17)} * sqrt(2/17) * 2 * sqrt(2)
+ // at precision of 2 * DCT_CONST_BITS bits
+ 47852167, 94074787, 137093803, 175444254,
+ 207820161, 233119001, 250479254, 259309736
+};
+#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
+
static INLINE tran_low_t check_range(tran_high_t input) {
#if CONFIG_VP9_HIGHBITDEPTH
// For valid highbitdepth VP9 streams, intermediate stage coefficients will
{default_scan_4x4, vp9_default_iscan_4x4, default_scan_4x4_neighbors},
#if CONFIG_DST1
{default_scan_4x4, vp9_default_iscan_4x4, default_scan_4x4_neighbors},
+ {default_scan_4x4, vp9_default_iscan_4x4, default_scan_4x4_neighbors},
+ {default_scan_4x4, vp9_default_iscan_4x4, default_scan_4x4_neighbors},
+ {default_scan_4x4, vp9_default_iscan_4x4, default_scan_4x4_neighbors},
+ {default_scan_4x4, vp9_default_iscan_4x4, default_scan_4x4_neighbors},
+ {default_scan_4x4, vp9_default_iscan_4x4, default_scan_4x4_neighbors},
+ {default_scan_4x4, vp9_default_iscan_4x4, default_scan_4x4_neighbors},
#endif // CONFIG_DST1
#if CONFIG_WAVELETS
{default_scan_4x4, vp9_default_iscan_4x4, default_scan_4x4_neighbors},
{default_scan_8x8, vp9_default_iscan_8x8, default_scan_8x8_neighbors},
#if CONFIG_DST1
{default_scan_8x8, vp9_default_iscan_8x8, default_scan_8x8_neighbors},
+ {default_scan_8x8, vp9_default_iscan_8x8, default_scan_8x8_neighbors},
+ {default_scan_8x8, vp9_default_iscan_8x8, default_scan_8x8_neighbors},
+ {default_scan_8x8, vp9_default_iscan_8x8, default_scan_8x8_neighbors},
+ {default_scan_8x8, vp9_default_iscan_8x8, default_scan_8x8_neighbors},
+ {default_scan_8x8, vp9_default_iscan_8x8, default_scan_8x8_neighbors},
+ {default_scan_8x8, vp9_default_iscan_8x8, default_scan_8x8_neighbors},
#endif // CONFIG_DST1
#if CONFIG_WAVELETS
{default_scan_8x8, vp9_default_iscan_8x8, default_scan_8x8_neighbors}
{default_scan_16x16, vp9_default_iscan_16x16, default_scan_16x16_neighbors},
#if CONFIG_DST1
{default_scan_16x16, vp9_default_iscan_16x16, default_scan_16x16_neighbors},
+ {default_scan_16x16, vp9_default_iscan_16x16, default_scan_16x16_neighbors},
+ {default_scan_16x16, vp9_default_iscan_16x16, default_scan_16x16_neighbors},
+ {default_scan_16x16, vp9_default_iscan_16x16, default_scan_16x16_neighbors},
+ {default_scan_16x16, vp9_default_iscan_16x16, default_scan_16x16_neighbors},
+ {default_scan_16x16, vp9_default_iscan_16x16, default_scan_16x16_neighbors},
+ {default_scan_16x16, vp9_default_iscan_16x16, default_scan_16x16_neighbors},
#endif // CONFIG_DST1
#if CONFIG_WAVELETS
{default_scan_16x16, vp9_default_iscan_16x16, default_scan_16x16_neighbors}
{default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors},
#if CONFIG_DST1
{default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors},
+ {default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors},
+ {default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors},
+ {default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors},
+ {default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors},
+ {default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors},
+ {default_scan_32x32, vp9_default_iscan_32x32, default_scan_32x32_neighbors},
#endif // CONFIG_DST1
#if CONFIG_WAVELETS
{dwtdct_scan_32x32, vp9_dwtdct_iscan_32x32, dwtdct_scan_32x32_neighbors},
{default_scan_64x64, vp9_default_iscan_64x64, default_scan_64x64_neighbors},
#if CONFIG_DST1
{default_scan_64x64, vp9_default_iscan_64x64, default_scan_64x64_neighbors},
+ {default_scan_64x64, vp9_default_iscan_64x64, default_scan_64x64_neighbors},
+ {default_scan_64x64, vp9_default_iscan_64x64, default_scan_64x64_neighbors},
+ {default_scan_64x64, vp9_default_iscan_64x64, default_scan_64x64_neighbors},
+ {default_scan_64x64, vp9_default_iscan_64x64, default_scan_64x64_neighbors},
+ {default_scan_64x64, vp9_default_iscan_64x64, default_scan_64x64_neighbors},
+ {default_scan_64x64, vp9_default_iscan_64x64, default_scan_64x64_neighbors},
#endif // CONFIG_DST1
#if CONFIG_WAVELETS
{dwtdct_scan_64x64, vp9_dwtdct_iscan_64x64, dwtdct_scan_64x64_neighbors},
return rv;
}
+#if CONFIG_EXT_TX
#if CONFIG_DST1
-static void fdstNxN(const int16_t *input, tran_low_t *output,
- int stride, int N) {
- const int val_2_5 = 6554;
- const int val_2_9 = 3641;
- const int val_2_17 = 1928;
+static INLINE void vp9_fdst_core(const tran_low_t *input, tran_low_t *output,
+ const int N, const int32_t* dst_lookup) {
int i, j;
- int64_t *in = (int64_t *) malloc (N * sizeof(int64_t));
- int64_t *inter = (int64_t *) malloc (N * sizeof(int64_t));
- int64_t *mat = (int64_t *) malloc (N * N * sizeof(int64_t));
- int64_t *mat2 = (int64_t *) malloc (N * N * sizeof(int64_t));
- int64_t val;
-
- // 1d dst: transform columns
- for (j = 0; j < N; j++) {
- for (i = 0; i < N; i++) {
- in[i] = input[i * stride + j];
- }
- vp9_dst1d_type1(in, inter, N);
- for (i = 0; i < N; i++)
- mat2[i * N + j] = inter[i];
- }
- // transpose
- for (i = 0; i < N; i++)
- for (j = 0; j < N; j++)
- mat[i*N + j] = mat2[i + j*N];
-
- switch (N) {
- case 4:
- val = val_2_5;
- break;
- case 8:
- val = val_2_9;
- break;
- case 16:
- val = val_2_17;
- break;
- default:
- return;
- }
- // 1d dst: transform rows
- for (j = 0; j < N; j++) {
- for (i = 0; i < N; i++) {
- in[i] = mat[i * N + j];
- }
- vp9_dst1d_type1(in, inter, N);
- for (i = 0; i < N; i++) {
- int64_t tmp = inter[i];
- tmp = tmp >> DCT_CONST_BITS;
- tmp *= val;
- mat[i * N + j] = tmp >> (2 * DCT_CONST_BITS - 3);
- }
- }
for (i = 0; i < N; i++) {
+ int64_t sum = 0;
for (j = 0; j < N; j++) {
- output[i * N + j] = WRAPLOW(mat[i * N + j], 8);
+ int idx = (i + 1) * (j + 1);
+ int sign = 0;
+ if (idx > N + 1) {
+ sign = (idx / (N + 1)) & 1;
+ idx %= (N + 1);
+ }
+ idx = MIN(idx, N + 1 - idx);
+ if (idx == 0) continue;
+ idx--;
+ sum += (int64_t)input[j] * dst_lookup[idx] * (sign ? -1 : 1);
}
+ output[i] = ROUND_POWER_OF_TWO(sum, (2 * DCT_CONST_BITS));
}
- free(in);
- free(inter);
- free(mat);
- free(mat2);
}
-void vp9_fdst4x4(const int16_t *input, tran_low_t *output, int stride) {
- fdstNxN(input, output, stride, 4);
+void vp9_fdst4(const tran_low_t *input, tran_low_t *output) {
+ vp9_fdst_core(input, output, 4, dst_lookup4);
}
-void vp9_fdst8x8(const int16_t *input, tran_low_t *output, int stride) {
- fdstNxN(input, output, stride, 8);
+void vp9_fdst8(const tran_low_t *input, tran_low_t *output) {
+ vp9_fdst_core(input, output, 8, dst_lookup8);
}
-void vp9_fdst16x16(const int16_t *input, tran_low_t *output, int stride) {
- fdstNxN(input, output, stride, 16);
+void vp9_fdst16(const tran_low_t *input, tran_low_t *output) {
+ vp9_fdst_core(input, output, 16, dst_lookup16);
}
#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
void vp9_fdct4(const tran_low_t *input, tran_low_t *output) {
tran_high_t step[4];
void vp9_fadst16(const tran_low_t *input, tran_low_t *output);
void vp9_fdct32(const tran_high_t *input, tran_high_t *output, int round);
+#if CONFIG_EXT_TX
#if CONFIG_DST1
-void vp9_fdst4x4(const int16_t *input, tran_low_t *output, int stride);
-void vp9_fdst8x8(const int16_t *input, tran_low_t *output, int stride);
-void vp9_fdst16x16(const int16_t *input, tran_low_t *output, int stride);
+void vp9_fdst4(const tran_low_t *input, tran_low_t *output);
+void vp9_fdst8(const tran_low_t *input, tran_low_t *output);
+void vp9_fdst16(const tran_low_t *input, tran_low_t *output);
#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
static const transform_2d FHT_4[] = {
{ vp9_fdct4, vp9_fdct4 }, // DCT_DCT = 0
{ vp9_fadst4, vp9_fdct4 }, // ADST_DCT = 1
{ vp9_fdct4, vp9_fadst4 }, // DCT_ADST = 2
- { vp9_fadst4, vp9_fadst4 } // ADST_ADST = 3
+ { vp9_fadst4, vp9_fadst4 }, // ADST_ADST = 3
+#if CONFIG_EXT_TX
+ { vp9_fadst4, vp9_fdct4 }, // FLIPADST_DCT = 4
+ { vp9_fdct4, vp9_fadst4 }, // DCT_FLIPADST = 5
+ { vp9_fadst4, vp9_fadst4 }, // FLIPADST_FLIPADST = 6
+ { vp9_fadst4, vp9_fadst4 }, // ADST_FLIPADST = 7
+ { vp9_fadst4, vp9_fadst4 }, // FLIPADST_ADST = 8
+#if CONFIG_DST1
+ { vp9_fdst4, vp9_fdst4 }, // DST_DST = 9
+ { vp9_fdst4, vp9_fdct4 }, // DST_DCT = 10
+ { vp9_fdct4, vp9_fdst4 }, // DCT_DST = 11
+ { vp9_fdst4, vp9_fadst4 }, // DST_ADST = 12
+ { vp9_fadst4, vp9_fdst4 }, // ADST_DST = 13
+ { vp9_fdst4, vp9_fadst4 }, // DST_FLIPADST = 14
+ { vp9_fadst4, vp9_fdst4 }, // FLIPADST_DST = 15
+#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
};
static const transform_2d FHT_8[] = {
{ vp9_fdct8, vp9_fdct8 }, // DCT_DCT = 0
{ vp9_fadst8, vp9_fdct8 }, // ADST_DCT = 1
{ vp9_fdct8, vp9_fadst8 }, // DCT_ADST = 2
- { vp9_fadst8, vp9_fadst8 } // ADST_ADST = 3
+ { vp9_fadst8, vp9_fadst8 }, // ADST_ADST = 3
+#if CONFIG_EXT_TX
+ { vp9_fadst8, vp9_fdct8 }, // FLIPADST_DCT = 4
+ { vp9_fdct8, vp9_fadst8 }, // DCT_FLIPADST = 5
+ { vp9_fadst8, vp9_fadst8 }, // FLIPADST_FLIPADST = 6
+ { vp9_fadst8, vp9_fadst8 }, // ADST_FLIPADST = 7
+ { vp9_fadst8, vp9_fadst8 }, // FLIPADST_ADST = 8
+#if CONFIG_DST1
+ { vp9_fdst8, vp9_fdst8 }, // DST_DST = 9
+ { vp9_fdst8, vp9_fdct8 }, // DST_DCT = 10
+ { vp9_fdct8, vp9_fdst8 }, // DCT_DST = 11
+ { vp9_fdst8, vp9_fadst8 }, // DST_ADST = 12
+ { vp9_fadst8, vp9_fdst8 }, // ADST_DST = 13
+ { vp9_fdst8, vp9_fadst8 }, // DST_FLIPADST = 14
+ { vp9_fadst8, vp9_fdst8 }, // FLIPADST_DST = 15
+#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
};
static const transform_2d FHT_16[] = {
{ vp9_fdct16, vp9_fdct16 }, // DCT_DCT = 0
{ vp9_fadst16, vp9_fdct16 }, // ADST_DCT = 1
{ vp9_fdct16, vp9_fadst16 }, // DCT_ADST = 2
- { vp9_fadst16, vp9_fadst16 } // ADST_ADST = 3
+ { vp9_fadst16, vp9_fadst16 }, // ADST_ADST = 3
+#if CONFIG_EXT_TX
+ { vp9_fadst16, vp9_fdct16 }, // FLIPADST_DCT = 4
+ { vp9_fdct16, vp9_fadst16 }, // DCT_FLIPADST = 5
+ { vp9_fadst16, vp9_fadst16 }, // FLIPADST_FLIPADST = 6
+ { vp9_fadst16, vp9_fadst16 }, // ADST_FLIPADST = 7
+ { vp9_fadst16, vp9_fadst16 }, // FLIPADST_ADST = 8
+#if CONFIG_DST1
+ { vp9_fdst16, vp9_fdst16 }, // DST_DST = 9
+ { vp9_fdst16, vp9_fdct16 }, // DST_DCT = 10
+ { vp9_fdct16, vp9_fdst16 }, // DCT_DST = 11
+ { vp9_fdst16, vp9_fadst16 }, // DST_ADST = 12
+ { vp9_fadst16, vp9_fdst16 }, // ADST_DST = 13
+ { vp9_fdst16, vp9_fadst16 }, // DST_FLIPADST = 14
+ { vp9_fadst16, vp9_fdst16 }, // FLIPADST_DST = 15
+#endif // CONFIG_DST1
+#endif // CONFIG_EXT_TX
};
#ifdef __cplusplus
if (tx_type == DCT_DCT) {
vp9_fdct16x16(src_diff, coeff, diff_stride);
#if CONFIG_DST1
- } else if (tx_type == DST_DST) {
- vp9_fdst16x16(src_diff, coeff, diff_stride);
+ } else if (is_dst_used(tx_type)) {
+ vp9_fht16x16_c(src_diff, coeff, diff_stride, tx_type);
#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
copy_flipud(src_diff, diff_stride, 16, src_diff2, 16);
if (tx_type == DCT_DCT) {
vp9_fdct8x8(src_diff, coeff, diff_stride);
#if CONFIG_DST1
- } else if (tx_type == DST_DST) {
- vp9_fdst8x8(src_diff, coeff, diff_stride);
+ } else if (is_dst_used(tx_type)) {
+ vp9_fht8x8_c(src_diff, coeff, diff_stride, tx_type);
#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
copy_flipud(src_diff, diff_stride, 8, src_diff2, 8);
if (tx_type == DCT_DCT) {
x->fwd_txm4x4(src_diff, coeff, diff_stride);
#if CONFIG_DST1
- } else if (tx_type == DST_DST) {
- vp9_fdst4x4(src_diff, coeff, diff_stride);
+ } else if (is_dst_used(tx_type)) {
+ vp9_fht4x4_c(src_diff, coeff, diff_stride, tx_type);
#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
copy_flipud(src_diff, diff_stride, 4, src_diff2, 4);
TX_TYPE tx_type = get_tx_type(plane, xd);
if (tx_type == DCT_DCT) {
vp9_highbd_fdct16x16(src_diff, coeff, diff_stride);
+#if CONFIG_DST1
+ } else if (is_dst_used(tx_type)) {
+ vp9_highbd_fht16x16_c(src_diff, coeff, diff_stride, tx_type);
+#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
copy_flipud(src_diff, diff_stride, 16, src_diff2, 16);
vp9_highbd_fht16x16(src_diff2, coeff, 16, ADST_DCT);
TX_TYPE tx_type = get_tx_type(plane, xd);
if (tx_type == DCT_DCT) {
vp9_highbd_fdct8x8(src_diff, coeff, diff_stride);
+#if CONFIG_DST1
+ } else if (is_dst_used(tx_type)) {
+ vp9_highbd_fht8x8_c(src_diff, coeff, diff_stride, tx_type);
+#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
copy_flipud(src_diff, diff_stride, 8, src_diff2, 8);
vp9_highbd_fht8x8(src_diff2, coeff, 8, ADST_DCT);
TX_TYPE tx_type = get_tx_type_4x4(plane, xd, block);
if (tx_type == DCT_DCT) {
x->fwd_txm4x4(src_diff, coeff, diff_stride);
+#if CONFIG_DST1
+ } else if (is_dst_used(tx_type)) {
+ vp9_highbd_fht4x4_c(src_diff, coeff, diff_stride, tx_type);
+#endif // CONFIG_DST1
} else if (tx_type == FLIPADST_DCT) {
copy_flipud(src_diff, diff_stride, 4, src_diff2, 4);
vp9_highbd_fht4x4(src_diff2, coeff, 4, ADST_DCT);
#define MIN_EARLY_TERM_INDEX 3
#if CONFIG_EXT_TX
-const double ext_tx_th = 0.99;
+const double ext_tx_th = 0.98;
#endif
typedef struct {
} else if (s[n]) {
if (is_inter_block(mbmi)) {
rd[n][0] = rd[n][1] = RDCOST(x->rdmult, x->rddiv, s1, sse[n]);
+ r[n][1] -= r_tx_size;
} else {
#if CONFIG_SR_MODE
rd[n][0] = RDCOST(x->rdmult, x->rddiv, s1 + sr0, sse[n]);
int64_t distortion_y_tx;
int dummy;
int64_t best_rdcost_tx = INT64_MAX;
- int best_ext_tx = NORM;
+ int best_ext_tx = -1;
for (i = NORM; i < EXT_TX_TYPES; i++) {
mbmi->ext_txfrm = i;
rdcost_tx = RDCOST(x->rdmult, x->rddiv, rate_y_tx, distortion_y_tx);
rdcost_tx = MIN(rdcost_tx, RDCOST(x->rdmult, x->rddiv, 0, *psse));
assert(rdcost_tx >= 0);
- if (rdcost_tx < best_rdcost_tx * ext_tx_th) {
+ if (rdcost_tx <
+ best_rdcost_tx * (best_ext_tx == NORM ? ext_tx_th : 1)) {
best_ext_tx = i;
best_rdcost_tx = rdcost_tx;
}
}
#if CONFIG_EXT_TX
this_rd = RDCOST(x->rdmult, x->rddiv, rate2, distortion2);
- if (tx_type == NORM || this_rd < (bestrd_tx * ext_tx_th)) {
+ if (tx_type == NORM ||
+ this_rd < bestrd_tx * (best_tx_type == NORM ? ext_tx_th : 1.0)) {
bestrd_tx = this_rd;
best_tx_type = tx_type;
best_tx_size = mbmi->tx_size;
projects / libvpx / commitdiff