1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp9/encoder/vp9_dct.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,1401 @@
1.4 +/*
1.5 + * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
1.6 + *
1.7 + * Use of this source code is governed by a BSD-style license
1.8 + * that can be found in the LICENSE file in the root of the source
1.9 + * tree. An additional intellectual property rights grant can be found
1.10 + * in the file PATENTS. All contributing project authors may
1.11 + * be found in the AUTHORS file in the root of the source tree.
1.12 + */
1.13 +
1.14 +#include <assert.h>
1.15 +#include <math.h>
1.16 +
1.17 +#include "./vpx_config.h"
1.18 +#include "./vp9_rtcd.h"
1.19 +
1.20 +#include "vp9/common/vp9_blockd.h"
1.21 +#include "vp9/common/vp9_idct.h"
1.22 +#include "vp9/common/vp9_systemdependent.h"
1.23 +
1.24 +#include "vp9/encoder/vp9_dct.h"
1.25 +
1.26 +static INLINE int fdct_round_shift(int input) {
1.27 + int rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
1.28 + assert(INT16_MIN <= rv && rv <= INT16_MAX);
1.29 + return rv;
1.30 +}
1.31 +
1.32 +static void fdct4(const int16_t *input, int16_t *output) {
1.33 + int16_t step[4];
1.34 + int temp1, temp2;
1.35 +
1.36 + step[0] = input[0] + input[3];
1.37 + step[1] = input[1] + input[2];
1.38 + step[2] = input[1] - input[2];
1.39 + step[3] = input[0] - input[3];
1.40 +
1.41 + temp1 = (step[0] + step[1]) * cospi_16_64;
1.42 + temp2 = (step[0] - step[1]) * cospi_16_64;
1.43 + output[0] = fdct_round_shift(temp1);
1.44 + output[2] = fdct_round_shift(temp2);
1.45 + temp1 = step[2] * cospi_24_64 + step[3] * cospi_8_64;
1.46 + temp2 = -step[2] * cospi_8_64 + step[3] * cospi_24_64;
1.47 + output[1] = fdct_round_shift(temp1);
1.48 + output[3] = fdct_round_shift(temp2);
1.49 +}
1.50 +
1.51 +void vp9_fdct4x4_c(const int16_t *input, int16_t *output, int stride) {
1.52 + // The 2D transform is done with two passes which are actually pretty
1.53 + // similar. In the first one, we transform the columns and transpose
1.54 + // the results. In the second one, we transform the rows. To achieve that,
1.55 + // as the first pass results are transposed, we tranpose the columns (that
1.56 + // is the transposed rows) and transpose the results (so that it goes back
1.57 + // in normal/row positions).
1.58 + int pass;
1.59 + // We need an intermediate buffer between passes.
1.60 + int16_t intermediate[4 * 4];
1.61 + const int16_t *in = input;
1.62 + int16_t *out = intermediate;
1.63 + // Do the two transform/transpose passes
1.64 + for (pass = 0; pass < 2; ++pass) {
1.65 + /*canbe16*/ int input[4];
1.66 + /*canbe16*/ int step[4];
1.67 + /*needs32*/ int temp1, temp2;
1.68 + int i;
1.69 + for (i = 0; i < 4; ++i) {
1.70 + // Load inputs.
1.71 + if (0 == pass) {
1.72 + input[0] = in[0 * stride] * 16;
1.73 + input[1] = in[1 * stride] * 16;
1.74 + input[2] = in[2 * stride] * 16;
1.75 + input[3] = in[3 * stride] * 16;
1.76 + if (i == 0 && input[0]) {
1.77 + input[0] += 1;
1.78 + }
1.79 + } else {
1.80 + input[0] = in[0 * 4];
1.81 + input[1] = in[1 * 4];
1.82 + input[2] = in[2 * 4];
1.83 + input[3] = in[3 * 4];
1.84 + }
1.85 + // Transform.
1.86 + step[0] = input[0] + input[3];
1.87 + step[1] = input[1] + input[2];
1.88 + step[2] = input[1] - input[2];
1.89 + step[3] = input[0] - input[3];
1.90 + temp1 = (step[0] + step[1]) * cospi_16_64;
1.91 + temp2 = (step[0] - step[1]) * cospi_16_64;
1.92 + out[0] = fdct_round_shift(temp1);
1.93 + out[2] = fdct_round_shift(temp2);
1.94 + temp1 = step[2] * cospi_24_64 + step[3] * cospi_8_64;
1.95 + temp2 = -step[2] * cospi_8_64 + step[3] * cospi_24_64;
1.96 + out[1] = fdct_round_shift(temp1);
1.97 + out[3] = fdct_round_shift(temp2);
1.98 + // Do next column (which is a transposed row in second/horizontal pass)
1.99 + in++;
1.100 + out += 4;
1.101 + }
1.102 + // Setup in/out for next pass.
1.103 + in = intermediate;
1.104 + out = output;
1.105 + }
1.106 +
1.107 + {
1.108 + int i, j;
1.109 + for (i = 0; i < 4; ++i) {
1.110 + for (j = 0; j < 4; ++j)
1.111 + output[j + i * 4] = (output[j + i * 4] + 1) >> 2;
1.112 + }
1.113 + }
1.114 +}
1.115 +
1.116 +static void fadst4(const int16_t *input, int16_t *output) {
1.117 + int x0, x1, x2, x3;
1.118 + int s0, s1, s2, s3, s4, s5, s6, s7;
1.119 +
1.120 + x0 = input[0];
1.121 + x1 = input[1];
1.122 + x2 = input[2];
1.123 + x3 = input[3];
1.124 +
1.125 + if (!(x0 | x1 | x2 | x3)) {
1.126 + output[0] = output[1] = output[2] = output[3] = 0;
1.127 + return;
1.128 + }
1.129 +
1.130 + s0 = sinpi_1_9 * x0;
1.131 + s1 = sinpi_4_9 * x0;
1.132 + s2 = sinpi_2_9 * x1;
1.133 + s3 = sinpi_1_9 * x1;
1.134 + s4 = sinpi_3_9 * x2;
1.135 + s5 = sinpi_4_9 * x3;
1.136 + s6 = sinpi_2_9 * x3;
1.137 + s7 = x0 + x1 - x3;
1.138 +
1.139 + x0 = s0 + s2 + s5;
1.140 + x1 = sinpi_3_9 * s7;
1.141 + x2 = s1 - s3 + s6;
1.142 + x3 = s4;
1.143 +
1.144 + s0 = x0 + x3;
1.145 + s1 = x1;
1.146 + s2 = x2 - x3;
1.147 + s3 = x2 - x0 + x3;
1.148 +
1.149 + // 1-D transform scaling factor is sqrt(2).
1.150 + output[0] = fdct_round_shift(s0);
1.151 + output[1] = fdct_round_shift(s1);
1.152 + output[2] = fdct_round_shift(s2);
1.153 + output[3] = fdct_round_shift(s3);
1.154 +}
1.155 +
1.156 +static const transform_2d FHT_4[] = {
1.157 + { fdct4, fdct4 }, // DCT_DCT = 0
1.158 + { fadst4, fdct4 }, // ADST_DCT = 1
1.159 + { fdct4, fadst4 }, // DCT_ADST = 2
1.160 + { fadst4, fadst4 } // ADST_ADST = 3
1.161 +};
1.162 +
1.163 +void vp9_short_fht4x4_c(const int16_t *input, int16_t *output,
1.164 + int stride, int tx_type) {
1.165 + int16_t out[4 * 4];
1.166 + int16_t *outptr = &out[0];
1.167 + int i, j;
1.168 + int16_t temp_in[4], temp_out[4];
1.169 + const transform_2d ht = FHT_4[tx_type];
1.170 +
1.171 + // Columns
1.172 + for (i = 0; i < 4; ++i) {
1.173 + for (j = 0; j < 4; ++j)
1.174 + temp_in[j] = input[j * stride + i] * 16;
1.175 + if (i == 0 && temp_in[0])
1.176 + temp_in[0] += 1;
1.177 + ht.cols(temp_in, temp_out);
1.178 + for (j = 0; j < 4; ++j)
1.179 + outptr[j * 4 + i] = temp_out[j];
1.180 + }
1.181 +
1.182 + // Rows
1.183 + for (i = 0; i < 4; ++i) {
1.184 + for (j = 0; j < 4; ++j)
1.185 + temp_in[j] = out[j + i * 4];
1.186 + ht.rows(temp_in, temp_out);
1.187 + for (j = 0; j < 4; ++j)
1.188 + output[j + i * 4] = (temp_out[j] + 1) >> 2;
1.189 + }
1.190 +}
1.191 +
1.192 +static void fdct8(const int16_t *input, int16_t *output) {
1.193 + /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
1.194 + /*needs32*/ int t0, t1, t2, t3;
1.195 + /*canbe16*/ int x0, x1, x2, x3;
1.196 +
1.197 + // stage 1
1.198 + s0 = input[0] + input[7];
1.199 + s1 = input[1] + input[6];
1.200 + s2 = input[2] + input[5];
1.201 + s3 = input[3] + input[4];
1.202 + s4 = input[3] - input[4];
1.203 + s5 = input[2] - input[5];
1.204 + s6 = input[1] - input[6];
1.205 + s7 = input[0] - input[7];
1.206 +
1.207 + // fdct4(step, step);
1.208 + x0 = s0 + s3;
1.209 + x1 = s1 + s2;
1.210 + x2 = s1 - s2;
1.211 + x3 = s0 - s3;
1.212 + t0 = (x0 + x1) * cospi_16_64;
1.213 + t1 = (x0 - x1) * cospi_16_64;
1.214 + t2 = x2 * cospi_24_64 + x3 * cospi_8_64;
1.215 + t3 = -x2 * cospi_8_64 + x3 * cospi_24_64;
1.216 + output[0] = fdct_round_shift(t0);
1.217 + output[2] = fdct_round_shift(t2);
1.218 + output[4] = fdct_round_shift(t1);
1.219 + output[6] = fdct_round_shift(t3);
1.220 +
1.221 + // Stage 2
1.222 + t0 = (s6 - s5) * cospi_16_64;
1.223 + t1 = (s6 + s5) * cospi_16_64;
1.224 + t2 = fdct_round_shift(t0);
1.225 + t3 = fdct_round_shift(t1);
1.226 +
1.227 + // Stage 3
1.228 + x0 = s4 + t2;
1.229 + x1 = s4 - t2;
1.230 + x2 = s7 - t3;
1.231 + x3 = s7 + t3;
1.232 +
1.233 + // Stage 4
1.234 + t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
1.235 + t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
1.236 + t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
1.237 + t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
1.238 + output[1] = fdct_round_shift(t0);
1.239 + output[3] = fdct_round_shift(t2);
1.240 + output[5] = fdct_round_shift(t1);
1.241 + output[7] = fdct_round_shift(t3);
1.242 +}
1.243 +
1.244 +void vp9_fdct8x8_c(const int16_t *input, int16_t *final_output, int stride) {
1.245 + int i, j;
1.246 + int16_t intermediate[64];
1.247 +
1.248 + // Transform columns
1.249 + {
1.250 + int16_t *output = intermediate;
1.251 + /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
1.252 + /*needs32*/ int t0, t1, t2, t3;
1.253 + /*canbe16*/ int x0, x1, x2, x3;
1.254 +
1.255 + int i;
1.256 + for (i = 0; i < 8; i++) {
1.257 + // stage 1
1.258 + s0 = (input[0 * stride] + input[7 * stride]) * 4;
1.259 + s1 = (input[1 * stride] + input[6 * stride]) * 4;
1.260 + s2 = (input[2 * stride] + input[5 * stride]) * 4;
1.261 + s3 = (input[3 * stride] + input[4 * stride]) * 4;
1.262 + s4 = (input[3 * stride] - input[4 * stride]) * 4;
1.263 + s5 = (input[2 * stride] - input[5 * stride]) * 4;
1.264 + s6 = (input[1 * stride] - input[6 * stride]) * 4;
1.265 + s7 = (input[0 * stride] - input[7 * stride]) * 4;
1.266 +
1.267 + // fdct4(step, step);
1.268 + x0 = s0 + s3;
1.269 + x1 = s1 + s2;
1.270 + x2 = s1 - s2;
1.271 + x3 = s0 - s3;
1.272 + t0 = (x0 + x1) * cospi_16_64;
1.273 + t1 = (x0 - x1) * cospi_16_64;
1.274 + t2 = x2 * cospi_24_64 + x3 * cospi_8_64;
1.275 + t3 = -x2 * cospi_8_64 + x3 * cospi_24_64;
1.276 + output[0 * 8] = fdct_round_shift(t0);
1.277 + output[2 * 8] = fdct_round_shift(t2);
1.278 + output[4 * 8] = fdct_round_shift(t1);
1.279 + output[6 * 8] = fdct_round_shift(t3);
1.280 +
1.281 + // Stage 2
1.282 + t0 = (s6 - s5) * cospi_16_64;
1.283 + t1 = (s6 + s5) * cospi_16_64;
1.284 + t2 = fdct_round_shift(t0);
1.285 + t3 = fdct_round_shift(t1);
1.286 +
1.287 + // Stage 3
1.288 + x0 = s4 + t2;
1.289 + x1 = s4 - t2;
1.290 + x2 = s7 - t3;
1.291 + x3 = s7 + t3;
1.292 +
1.293 + // Stage 4
1.294 + t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
1.295 + t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
1.296 + t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
1.297 + t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
1.298 + output[1 * 8] = fdct_round_shift(t0);
1.299 + output[3 * 8] = fdct_round_shift(t2);
1.300 + output[5 * 8] = fdct_round_shift(t1);
1.301 + output[7 * 8] = fdct_round_shift(t3);
1.302 + input++;
1.303 + output++;
1.304 + }
1.305 + }
1.306 +
1.307 + // Rows
1.308 + for (i = 0; i < 8; ++i) {
1.309 + fdct8(&intermediate[i * 8], &final_output[i * 8]);
1.310 + for (j = 0; j < 8; ++j)
1.311 + final_output[j + i * 8] /= 2;
1.312 + }
1.313 +}
1.314 +
1.315 +void vp9_fdct16x16_c(const int16_t *input, int16_t *output, int stride) {
1.316 + // The 2D transform is done with two passes which are actually pretty
1.317 + // similar. In the first one, we transform the columns and transpose
1.318 + // the results. In the second one, we transform the rows. To achieve that,
1.319 + // as the first pass results are transposed, we tranpose the columns (that
1.320 + // is the transposed rows) and transpose the results (so that it goes back
1.321 + // in normal/row positions).
1.322 + int pass;
1.323 + // We need an intermediate buffer between passes.
1.324 + int16_t intermediate[256];
1.325 + const int16_t *in = input;
1.326 + int16_t *out = intermediate;
1.327 + // Do the two transform/transpose passes
1.328 + for (pass = 0; pass < 2; ++pass) {
1.329 + /*canbe16*/ int step1[8];
1.330 + /*canbe16*/ int step2[8];
1.331 + /*canbe16*/ int step3[8];
1.332 + /*canbe16*/ int input[8];
1.333 + /*needs32*/ int temp1, temp2;
1.334 + int i;
1.335 + for (i = 0; i < 16; i++) {
1.336 + if (0 == pass) {
1.337 + // Calculate input for the first 8 results.
1.338 + input[0] = (in[0 * stride] + in[15 * stride]) * 4;
1.339 + input[1] = (in[1 * stride] + in[14 * stride]) * 4;
1.340 + input[2] = (in[2 * stride] + in[13 * stride]) * 4;
1.341 + input[3] = (in[3 * stride] + in[12 * stride]) * 4;
1.342 + input[4] = (in[4 * stride] + in[11 * stride]) * 4;
1.343 + input[5] = (in[5 * stride] + in[10 * stride]) * 4;
1.344 + input[6] = (in[6 * stride] + in[ 9 * stride]) * 4;
1.345 + input[7] = (in[7 * stride] + in[ 8 * stride]) * 4;
1.346 + // Calculate input for the next 8 results.
1.347 + step1[0] = (in[7 * stride] - in[ 8 * stride]) * 4;
1.348 + step1[1] = (in[6 * stride] - in[ 9 * stride]) * 4;
1.349 + step1[2] = (in[5 * stride] - in[10 * stride]) * 4;
1.350 + step1[3] = (in[4 * stride] - in[11 * stride]) * 4;
1.351 + step1[4] = (in[3 * stride] - in[12 * stride]) * 4;
1.352 + step1[5] = (in[2 * stride] - in[13 * stride]) * 4;
1.353 + step1[6] = (in[1 * stride] - in[14 * stride]) * 4;
1.354 + step1[7] = (in[0 * stride] - in[15 * stride]) * 4;
1.355 + } else {
1.356 + // Calculate input for the first 8 results.
1.357 + input[0] = ((in[0 * 16] + 1) >> 2) + ((in[15 * 16] + 1) >> 2);
1.358 + input[1] = ((in[1 * 16] + 1) >> 2) + ((in[14 * 16] + 1) >> 2);
1.359 + input[2] = ((in[2 * 16] + 1) >> 2) + ((in[13 * 16] + 1) >> 2);
1.360 + input[3] = ((in[3 * 16] + 1) >> 2) + ((in[12 * 16] + 1) >> 2);
1.361 + input[4] = ((in[4 * 16] + 1) >> 2) + ((in[11 * 16] + 1) >> 2);
1.362 + input[5] = ((in[5 * 16] + 1) >> 2) + ((in[10 * 16] + 1) >> 2);
1.363 + input[6] = ((in[6 * 16] + 1) >> 2) + ((in[ 9 * 16] + 1) >> 2);
1.364 + input[7] = ((in[7 * 16] + 1) >> 2) + ((in[ 8 * 16] + 1) >> 2);
1.365 + // Calculate input for the next 8 results.
1.366 + step1[0] = ((in[7 * 16] + 1) >> 2) - ((in[ 8 * 16] + 1) >> 2);
1.367 + step1[1] = ((in[6 * 16] + 1) >> 2) - ((in[ 9 * 16] + 1) >> 2);
1.368 + step1[2] = ((in[5 * 16] + 1) >> 2) - ((in[10 * 16] + 1) >> 2);
1.369 + step1[3] = ((in[4 * 16] + 1) >> 2) - ((in[11 * 16] + 1) >> 2);
1.370 + step1[4] = ((in[3 * 16] + 1) >> 2) - ((in[12 * 16] + 1) >> 2);
1.371 + step1[5] = ((in[2 * 16] + 1) >> 2) - ((in[13 * 16] + 1) >> 2);
1.372 + step1[6] = ((in[1 * 16] + 1) >> 2) - ((in[14 * 16] + 1) >> 2);
1.373 + step1[7] = ((in[0 * 16] + 1) >> 2) - ((in[15 * 16] + 1) >> 2);
1.374 + }
1.375 + // Work on the first eight values; fdct8(input, even_results);
1.376 + {
1.377 + /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
1.378 + /*needs32*/ int t0, t1, t2, t3;
1.379 + /*canbe16*/ int x0, x1, x2, x3;
1.380 +
1.381 + // stage 1
1.382 + s0 = input[0] + input[7];
1.383 + s1 = input[1] + input[6];
1.384 + s2 = input[2] + input[5];
1.385 + s3 = input[3] + input[4];
1.386 + s4 = input[3] - input[4];
1.387 + s5 = input[2] - input[5];
1.388 + s6 = input[1] - input[6];
1.389 + s7 = input[0] - input[7];
1.390 +
1.391 + // fdct4(step, step);
1.392 + x0 = s0 + s3;
1.393 + x1 = s1 + s2;
1.394 + x2 = s1 - s2;
1.395 + x3 = s0 - s3;
1.396 + t0 = (x0 + x1) * cospi_16_64;
1.397 + t1 = (x0 - x1) * cospi_16_64;
1.398 + t2 = x3 * cospi_8_64 + x2 * cospi_24_64;
1.399 + t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
1.400 + out[0] = fdct_round_shift(t0);
1.401 + out[4] = fdct_round_shift(t2);
1.402 + out[8] = fdct_round_shift(t1);
1.403 + out[12] = fdct_round_shift(t3);
1.404 +
1.405 + // Stage 2
1.406 + t0 = (s6 - s5) * cospi_16_64;
1.407 + t1 = (s6 + s5) * cospi_16_64;
1.408 + t2 = fdct_round_shift(t0);
1.409 + t3 = fdct_round_shift(t1);
1.410 +
1.411 + // Stage 3
1.412 + x0 = s4 + t2;
1.413 + x1 = s4 - t2;
1.414 + x2 = s7 - t3;
1.415 + x3 = s7 + t3;
1.416 +
1.417 + // Stage 4
1.418 + t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
1.419 + t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
1.420 + t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
1.421 + t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
1.422 + out[2] = fdct_round_shift(t0);
1.423 + out[6] = fdct_round_shift(t2);
1.424 + out[10] = fdct_round_shift(t1);
1.425 + out[14] = fdct_round_shift(t3);
1.426 + }
1.427 + // Work on the next eight values; step1 -> odd_results
1.428 + {
1.429 + // step 2
1.430 + temp1 = (step1[5] - step1[2]) * cospi_16_64;
1.431 + temp2 = (step1[4] - step1[3]) * cospi_16_64;
1.432 + step2[2] = fdct_round_shift(temp1);
1.433 + step2[3] = fdct_round_shift(temp2);
1.434 + temp1 = (step1[4] + step1[3]) * cospi_16_64;
1.435 + temp2 = (step1[5] + step1[2]) * cospi_16_64;
1.436 + step2[4] = fdct_round_shift(temp1);
1.437 + step2[5] = fdct_round_shift(temp2);
1.438 + // step 3
1.439 + step3[0] = step1[0] + step2[3];
1.440 + step3[1] = step1[1] + step2[2];
1.441 + step3[2] = step1[1] - step2[2];
1.442 + step3[3] = step1[0] - step2[3];
1.443 + step3[4] = step1[7] - step2[4];
1.444 + step3[5] = step1[6] - step2[5];
1.445 + step3[6] = step1[6] + step2[5];
1.446 + step3[7] = step1[7] + step2[4];
1.447 + // step 4
1.448 + temp1 = step3[1] * -cospi_8_64 + step3[6] * cospi_24_64;
1.449 + temp2 = step3[2] * -cospi_24_64 - step3[5] * cospi_8_64;
1.450 + step2[1] = fdct_round_shift(temp1);
1.451 + step2[2] = fdct_round_shift(temp2);
1.452 + temp1 = step3[2] * -cospi_8_64 + step3[5] * cospi_24_64;
1.453 + temp2 = step3[1] * cospi_24_64 + step3[6] * cospi_8_64;
1.454 + step2[5] = fdct_round_shift(temp1);
1.455 + step2[6] = fdct_round_shift(temp2);
1.456 + // step 5
1.457 + step1[0] = step3[0] + step2[1];
1.458 + step1[1] = step3[0] - step2[1];
1.459 + step1[2] = step3[3] - step2[2];
1.460 + step1[3] = step3[3] + step2[2];
1.461 + step1[4] = step3[4] + step2[5];
1.462 + step1[5] = step3[4] - step2[5];
1.463 + step1[6] = step3[7] - step2[6];
1.464 + step1[7] = step3[7] + step2[6];
1.465 + // step 6
1.466 + temp1 = step1[0] * cospi_30_64 + step1[7] * cospi_2_64;
1.467 + temp2 = step1[1] * cospi_14_64 + step1[6] * cospi_18_64;
1.468 + out[1] = fdct_round_shift(temp1);
1.469 + out[9] = fdct_round_shift(temp2);
1.470 + temp1 = step1[2] * cospi_22_64 + step1[5] * cospi_10_64;
1.471 + temp2 = step1[3] * cospi_6_64 + step1[4] * cospi_26_64;
1.472 + out[5] = fdct_round_shift(temp1);
1.473 + out[13] = fdct_round_shift(temp2);
1.474 + temp1 = step1[3] * -cospi_26_64 + step1[4] * cospi_6_64;
1.475 + temp2 = step1[2] * -cospi_10_64 + step1[5] * cospi_22_64;
1.476 + out[3] = fdct_round_shift(temp1);
1.477 + out[11] = fdct_round_shift(temp2);
1.478 + temp1 = step1[1] * -cospi_18_64 + step1[6] * cospi_14_64;
1.479 + temp2 = step1[0] * -cospi_2_64 + step1[7] * cospi_30_64;
1.480 + out[7] = fdct_round_shift(temp1);
1.481 + out[15] = fdct_round_shift(temp2);
1.482 + }
1.483 + // Do next column (which is a transposed row in second/horizontal pass)
1.484 + in++;
1.485 + out += 16;
1.486 + }
1.487 + // Setup in/out for next pass.
1.488 + in = intermediate;
1.489 + out = output;
1.490 + }
1.491 +}
1.492 +
1.493 +static void fadst8(const int16_t *input, int16_t *output) {
1.494 + int s0, s1, s2, s3, s4, s5, s6, s7;
1.495 +
1.496 + int x0 = input[7];
1.497 + int x1 = input[0];
1.498 + int x2 = input[5];
1.499 + int x3 = input[2];
1.500 + int x4 = input[3];
1.501 + int x5 = input[4];
1.502 + int x6 = input[1];
1.503 + int x7 = input[6];
1.504 +
1.505 + // stage 1
1.506 + s0 = cospi_2_64 * x0 + cospi_30_64 * x1;
1.507 + s1 = cospi_30_64 * x0 - cospi_2_64 * x1;
1.508 + s2 = cospi_10_64 * x2 + cospi_22_64 * x3;
1.509 + s3 = cospi_22_64 * x2 - cospi_10_64 * x3;
1.510 + s4 = cospi_18_64 * x4 + cospi_14_64 * x5;
1.511 + s5 = cospi_14_64 * x4 - cospi_18_64 * x5;
1.512 + s6 = cospi_26_64 * x6 + cospi_6_64 * x7;
1.513 + s7 = cospi_6_64 * x6 - cospi_26_64 * x7;
1.514 +
1.515 + x0 = fdct_round_shift(s0 + s4);
1.516 + x1 = fdct_round_shift(s1 + s5);
1.517 + x2 = fdct_round_shift(s2 + s6);
1.518 + x3 = fdct_round_shift(s3 + s7);
1.519 + x4 = fdct_round_shift(s0 - s4);
1.520 + x5 = fdct_round_shift(s1 - s5);
1.521 + x6 = fdct_round_shift(s2 - s6);
1.522 + x7 = fdct_round_shift(s3 - s7);
1.523 +
1.524 + // stage 2
1.525 + s0 = x0;
1.526 + s1 = x1;
1.527 + s2 = x2;
1.528 + s3 = x3;
1.529 + s4 = cospi_8_64 * x4 + cospi_24_64 * x5;
1.530 + s5 = cospi_24_64 * x4 - cospi_8_64 * x5;
1.531 + s6 = - cospi_24_64 * x6 + cospi_8_64 * x7;
1.532 + s7 = cospi_8_64 * x6 + cospi_24_64 * x7;
1.533 +
1.534 + x0 = s0 + s2;
1.535 + x1 = s1 + s3;
1.536 + x2 = s0 - s2;
1.537 + x3 = s1 - s3;
1.538 + x4 = fdct_round_shift(s4 + s6);
1.539 + x5 = fdct_round_shift(s5 + s7);
1.540 + x6 = fdct_round_shift(s4 - s6);
1.541 + x7 = fdct_round_shift(s5 - s7);
1.542 +
1.543 + // stage 3
1.544 + s2 = cospi_16_64 * (x2 + x3);
1.545 + s3 = cospi_16_64 * (x2 - x3);
1.546 + s6 = cospi_16_64 * (x6 + x7);
1.547 + s7 = cospi_16_64 * (x6 - x7);
1.548 +
1.549 + x2 = fdct_round_shift(s2);
1.550 + x3 = fdct_round_shift(s3);
1.551 + x6 = fdct_round_shift(s6);
1.552 + x7 = fdct_round_shift(s7);
1.553 +
1.554 + output[0] = x0;
1.555 + output[1] = - x4;
1.556 + output[2] = x6;
1.557 + output[3] = - x2;
1.558 + output[4] = x3;
1.559 + output[5] = - x7;
1.560 + output[6] = x5;
1.561 + output[7] = - x1;
1.562 +}
1.563 +
1.564 +static const transform_2d FHT_8[] = {
1.565 + { fdct8, fdct8 }, // DCT_DCT = 0
1.566 + { fadst8, fdct8 }, // ADST_DCT = 1
1.567 + { fdct8, fadst8 }, // DCT_ADST = 2
1.568 + { fadst8, fadst8 } // ADST_ADST = 3
1.569 +};
1.570 +
1.571 +void vp9_short_fht8x8_c(const int16_t *input, int16_t *output,
1.572 + int stride, int tx_type) {
1.573 + int16_t out[64];
1.574 + int16_t *outptr = &out[0];
1.575 + int i, j;
1.576 + int16_t temp_in[8], temp_out[8];
1.577 + const transform_2d ht = FHT_8[tx_type];
1.578 +
1.579 + // Columns
1.580 + for (i = 0; i < 8; ++i) {
1.581 + for (j = 0; j < 8; ++j)
1.582 + temp_in[j] = input[j * stride + i] * 4;
1.583 + ht.cols(temp_in, temp_out);
1.584 + for (j = 0; j < 8; ++j)
1.585 + outptr[j * 8 + i] = temp_out[j];
1.586 + }
1.587 +
1.588 + // Rows
1.589 + for (i = 0; i < 8; ++i) {
1.590 + for (j = 0; j < 8; ++j)
1.591 + temp_in[j] = out[j + i * 8];
1.592 + ht.rows(temp_in, temp_out);
1.593 + for (j = 0; j < 8; ++j)
1.594 + output[j + i * 8] = (temp_out[j] + (temp_out[j] < 0)) >> 1;
1.595 + }
1.596 +}
1.597 +
1.598 +/* 4-point reversible, orthonormal Walsh-Hadamard in 3.5 adds, 0.5 shifts per
1.599 + pixel. */
1.600 +void vp9_fwht4x4_c(const int16_t *input, int16_t *output, int stride) {
1.601 + int i;
1.602 + int a1, b1, c1, d1, e1;
1.603 + const int16_t *ip = input;
1.604 + int16_t *op = output;
1.605 +
1.606 + for (i = 0; i < 4; i++) {
1.607 + a1 = ip[0 * stride];
1.608 + b1 = ip[1 * stride];
1.609 + c1 = ip[2 * stride];
1.610 + d1 = ip[3 * stride];
1.611 +
1.612 + a1 += b1;
1.613 + d1 = d1 - c1;
1.614 + e1 = (a1 - d1) >> 1;
1.615 + b1 = e1 - b1;
1.616 + c1 = e1 - c1;
1.617 + a1 -= c1;
1.618 + d1 += b1;
1.619 + op[0] = a1;
1.620 + op[4] = c1;
1.621 + op[8] = d1;
1.622 + op[12] = b1;
1.623 +
1.624 + ip++;
1.625 + op++;
1.626 + }
1.627 + ip = output;
1.628 + op = output;
1.629 +
1.630 + for (i = 0; i < 4; i++) {
1.631 + a1 = ip[0];
1.632 + b1 = ip[1];
1.633 + c1 = ip[2];
1.634 + d1 = ip[3];
1.635 +
1.636 + a1 += b1;
1.637 + d1 -= c1;
1.638 + e1 = (a1 - d1) >> 1;
1.639 + b1 = e1 - b1;
1.640 + c1 = e1 - c1;
1.641 + a1 -= c1;
1.642 + d1 += b1;
1.643 + op[0] = a1 * UNIT_QUANT_FACTOR;
1.644 + op[1] = c1 * UNIT_QUANT_FACTOR;
1.645 + op[2] = d1 * UNIT_QUANT_FACTOR;
1.646 + op[3] = b1 * UNIT_QUANT_FACTOR;
1.647 +
1.648 + ip += 4;
1.649 + op += 4;
1.650 + }
1.651 +}
1.652 +
1.653 +// Rewrote to use same algorithm as others.
1.654 +static void fdct16(const int16_t in[16], int16_t out[16]) {
1.655 + /*canbe16*/ int step1[8];
1.656 + /*canbe16*/ int step2[8];
1.657 + /*canbe16*/ int step3[8];
1.658 + /*canbe16*/ int input[8];
1.659 + /*needs32*/ int temp1, temp2;
1.660 +
1.661 + // step 1
1.662 + input[0] = in[0] + in[15];
1.663 + input[1] = in[1] + in[14];
1.664 + input[2] = in[2] + in[13];
1.665 + input[3] = in[3] + in[12];
1.666 + input[4] = in[4] + in[11];
1.667 + input[5] = in[5] + in[10];
1.668 + input[6] = in[6] + in[ 9];
1.669 + input[7] = in[7] + in[ 8];
1.670 +
1.671 + step1[0] = in[7] - in[ 8];
1.672 + step1[1] = in[6] - in[ 9];
1.673 + step1[2] = in[5] - in[10];
1.674 + step1[3] = in[4] - in[11];
1.675 + step1[4] = in[3] - in[12];
1.676 + step1[5] = in[2] - in[13];
1.677 + step1[6] = in[1] - in[14];
1.678 + step1[7] = in[0] - in[15];
1.679 +
1.680 + // fdct8(step, step);
1.681 + {
1.682 + /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
1.683 + /*needs32*/ int t0, t1, t2, t3;
1.684 + /*canbe16*/ int x0, x1, x2, x3;
1.685 +
1.686 + // stage 1
1.687 + s0 = input[0] + input[7];
1.688 + s1 = input[1] + input[6];
1.689 + s2 = input[2] + input[5];
1.690 + s3 = input[3] + input[4];
1.691 + s4 = input[3] - input[4];
1.692 + s5 = input[2] - input[5];
1.693 + s6 = input[1] - input[6];
1.694 + s7 = input[0] - input[7];
1.695 +
1.696 + // fdct4(step, step);
1.697 + x0 = s0 + s3;
1.698 + x1 = s1 + s2;
1.699 + x2 = s1 - s2;
1.700 + x3 = s0 - s3;
1.701 + t0 = (x0 + x1) * cospi_16_64;
1.702 + t1 = (x0 - x1) * cospi_16_64;
1.703 + t2 = x3 * cospi_8_64 + x2 * cospi_24_64;
1.704 + t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
1.705 + out[0] = fdct_round_shift(t0);
1.706 + out[4] = fdct_round_shift(t2);
1.707 + out[8] = fdct_round_shift(t1);
1.708 + out[12] = fdct_round_shift(t3);
1.709 +
1.710 + // Stage 2
1.711 + t0 = (s6 - s5) * cospi_16_64;
1.712 + t1 = (s6 + s5) * cospi_16_64;
1.713 + t2 = fdct_round_shift(t0);
1.714 + t3 = fdct_round_shift(t1);
1.715 +
1.716 + // Stage 3
1.717 + x0 = s4 + t2;
1.718 + x1 = s4 - t2;
1.719 + x2 = s7 - t3;
1.720 + x3 = s7 + t3;
1.721 +
1.722 + // Stage 4
1.723 + t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
1.724 + t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
1.725 + t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
1.726 + t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
1.727 + out[2] = fdct_round_shift(t0);
1.728 + out[6] = fdct_round_shift(t2);
1.729 + out[10] = fdct_round_shift(t1);
1.730 + out[14] = fdct_round_shift(t3);
1.731 + }
1.732 +
1.733 + // step 2
1.734 + temp1 = (step1[5] - step1[2]) * cospi_16_64;
1.735 + temp2 = (step1[4] - step1[3]) * cospi_16_64;
1.736 + step2[2] = fdct_round_shift(temp1);
1.737 + step2[3] = fdct_round_shift(temp2);
1.738 + temp1 = (step1[4] + step1[3]) * cospi_16_64;
1.739 + temp2 = (step1[5] + step1[2]) * cospi_16_64;
1.740 + step2[4] = fdct_round_shift(temp1);
1.741 + step2[5] = fdct_round_shift(temp2);
1.742 +
1.743 + // step 3
1.744 + step3[0] = step1[0] + step2[3];
1.745 + step3[1] = step1[1] + step2[2];
1.746 + step3[2] = step1[1] - step2[2];
1.747 + step3[3] = step1[0] - step2[3];
1.748 + step3[4] = step1[7] - step2[4];
1.749 + step3[5] = step1[6] - step2[5];
1.750 + step3[6] = step1[6] + step2[5];
1.751 + step3[7] = step1[7] + step2[4];
1.752 +
1.753 + // step 4
1.754 + temp1 = step3[1] * -cospi_8_64 + step3[6] * cospi_24_64;
1.755 + temp2 = step3[2] * -cospi_24_64 - step3[5] * cospi_8_64;
1.756 + step2[1] = fdct_round_shift(temp1);
1.757 + step2[2] = fdct_round_shift(temp2);
1.758 + temp1 = step3[2] * -cospi_8_64 + step3[5] * cospi_24_64;
1.759 + temp2 = step3[1] * cospi_24_64 + step3[6] * cospi_8_64;
1.760 + step2[5] = fdct_round_shift(temp1);
1.761 + step2[6] = fdct_round_shift(temp2);
1.762 +
1.763 + // step 5
1.764 + step1[0] = step3[0] + step2[1];
1.765 + step1[1] = step3[0] - step2[1];
1.766 + step1[2] = step3[3] - step2[2];
1.767 + step1[3] = step3[3] + step2[2];
1.768 + step1[4] = step3[4] + step2[5];
1.769 + step1[5] = step3[4] - step2[5];
1.770 + step1[6] = step3[7] - step2[6];
1.771 + step1[7] = step3[7] + step2[6];
1.772 +
1.773 + // step 6
1.774 + temp1 = step1[0] * cospi_30_64 + step1[7] * cospi_2_64;
1.775 + temp2 = step1[1] * cospi_14_64 + step1[6] * cospi_18_64;
1.776 + out[1] = fdct_round_shift(temp1);
1.777 + out[9] = fdct_round_shift(temp2);
1.778 +
1.779 + temp1 = step1[2] * cospi_22_64 + step1[5] * cospi_10_64;
1.780 + temp2 = step1[3] * cospi_6_64 + step1[4] * cospi_26_64;
1.781 + out[5] = fdct_round_shift(temp1);
1.782 + out[13] = fdct_round_shift(temp2);
1.783 +
1.784 + temp1 = step1[3] * -cospi_26_64 + step1[4] * cospi_6_64;
1.785 + temp2 = step1[2] * -cospi_10_64 + step1[5] * cospi_22_64;
1.786 + out[3] = fdct_round_shift(temp1);
1.787 + out[11] = fdct_round_shift(temp2);
1.788 +
1.789 + temp1 = step1[1] * -cospi_18_64 + step1[6] * cospi_14_64;
1.790 + temp2 = step1[0] * -cospi_2_64 + step1[7] * cospi_30_64;
1.791 + out[7] = fdct_round_shift(temp1);
1.792 + out[15] = fdct_round_shift(temp2);
1.793 +}
1.794 +
1.795 +static void fadst16(const int16_t *input, int16_t *output) {
1.796 + int s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15;
1.797 +
1.798 + int x0 = input[15];
1.799 + int x1 = input[0];
1.800 + int x2 = input[13];
1.801 + int x3 = input[2];
1.802 + int x4 = input[11];
1.803 + int x5 = input[4];
1.804 + int x6 = input[9];
1.805 + int x7 = input[6];
1.806 + int x8 = input[7];
1.807 + int x9 = input[8];
1.808 + int x10 = input[5];
1.809 + int x11 = input[10];
1.810 + int x12 = input[3];
1.811 + int x13 = input[12];
1.812 + int x14 = input[1];
1.813 + int x15 = input[14];
1.814 +
1.815 + // stage 1
1.816 + s0 = x0 * cospi_1_64 + x1 * cospi_31_64;
1.817 + s1 = x0 * cospi_31_64 - x1 * cospi_1_64;
1.818 + s2 = x2 * cospi_5_64 + x3 * cospi_27_64;
1.819 + s3 = x2 * cospi_27_64 - x3 * cospi_5_64;
1.820 + s4 = x4 * cospi_9_64 + x5 * cospi_23_64;
1.821 + s5 = x4 * cospi_23_64 - x5 * cospi_9_64;
1.822 + s6 = x6 * cospi_13_64 + x7 * cospi_19_64;
1.823 + s7 = x6 * cospi_19_64 - x7 * cospi_13_64;
1.824 + s8 = x8 * cospi_17_64 + x9 * cospi_15_64;
1.825 + s9 = x8 * cospi_15_64 - x9 * cospi_17_64;
1.826 + s10 = x10 * cospi_21_64 + x11 * cospi_11_64;
1.827 + s11 = x10 * cospi_11_64 - x11 * cospi_21_64;
1.828 + s12 = x12 * cospi_25_64 + x13 * cospi_7_64;
1.829 + s13 = x12 * cospi_7_64 - x13 * cospi_25_64;
1.830 + s14 = x14 * cospi_29_64 + x15 * cospi_3_64;
1.831 + s15 = x14 * cospi_3_64 - x15 * cospi_29_64;
1.832 +
1.833 + x0 = fdct_round_shift(s0 + s8);
1.834 + x1 = fdct_round_shift(s1 + s9);
1.835 + x2 = fdct_round_shift(s2 + s10);
1.836 + x3 = fdct_round_shift(s3 + s11);
1.837 + x4 = fdct_round_shift(s4 + s12);
1.838 + x5 = fdct_round_shift(s5 + s13);
1.839 + x6 = fdct_round_shift(s6 + s14);
1.840 + x7 = fdct_round_shift(s7 + s15);
1.841 + x8 = fdct_round_shift(s0 - s8);
1.842 + x9 = fdct_round_shift(s1 - s9);
1.843 + x10 = fdct_round_shift(s2 - s10);
1.844 + x11 = fdct_round_shift(s3 - s11);
1.845 + x12 = fdct_round_shift(s4 - s12);
1.846 + x13 = fdct_round_shift(s5 - s13);
1.847 + x14 = fdct_round_shift(s6 - s14);
1.848 + x15 = fdct_round_shift(s7 - s15);
1.849 +
1.850 + // stage 2
1.851 + s0 = x0;
1.852 + s1 = x1;
1.853 + s2 = x2;
1.854 + s3 = x3;
1.855 + s4 = x4;
1.856 + s5 = x5;
1.857 + s6 = x6;
1.858 + s7 = x7;
1.859 + s8 = x8 * cospi_4_64 + x9 * cospi_28_64;
1.860 + s9 = x8 * cospi_28_64 - x9 * cospi_4_64;
1.861 + s10 = x10 * cospi_20_64 + x11 * cospi_12_64;
1.862 + s11 = x10 * cospi_12_64 - x11 * cospi_20_64;
1.863 + s12 = - x12 * cospi_28_64 + x13 * cospi_4_64;
1.864 + s13 = x12 * cospi_4_64 + x13 * cospi_28_64;
1.865 + s14 = - x14 * cospi_12_64 + x15 * cospi_20_64;
1.866 + s15 = x14 * cospi_20_64 + x15 * cospi_12_64;
1.867 +
1.868 + x0 = s0 + s4;
1.869 + x1 = s1 + s5;
1.870 + x2 = s2 + s6;
1.871 + x3 = s3 + s7;
1.872 + x4 = s0 - s4;
1.873 + x5 = s1 - s5;
1.874 + x6 = s2 - s6;
1.875 + x7 = s3 - s7;
1.876 + x8 = fdct_round_shift(s8 + s12);
1.877 + x9 = fdct_round_shift(s9 + s13);
1.878 + x10 = fdct_round_shift(s10 + s14);
1.879 + x11 = fdct_round_shift(s11 + s15);
1.880 + x12 = fdct_round_shift(s8 - s12);
1.881 + x13 = fdct_round_shift(s9 - s13);
1.882 + x14 = fdct_round_shift(s10 - s14);
1.883 + x15 = fdct_round_shift(s11 - s15);
1.884 +
1.885 + // stage 3
1.886 + s0 = x0;
1.887 + s1 = x1;
1.888 + s2 = x2;
1.889 + s3 = x3;
1.890 + s4 = x4 * cospi_8_64 + x5 * cospi_24_64;
1.891 + s5 = x4 * cospi_24_64 - x5 * cospi_8_64;
1.892 + s6 = - x6 * cospi_24_64 + x7 * cospi_8_64;
1.893 + s7 = x6 * cospi_8_64 + x7 * cospi_24_64;
1.894 + s8 = x8;
1.895 + s9 = x9;
1.896 + s10 = x10;
1.897 + s11 = x11;
1.898 + s12 = x12 * cospi_8_64 + x13 * cospi_24_64;
1.899 + s13 = x12 * cospi_24_64 - x13 * cospi_8_64;
1.900 + s14 = - x14 * cospi_24_64 + x15 * cospi_8_64;
1.901 + s15 = x14 * cospi_8_64 + x15 * cospi_24_64;
1.902 +
1.903 + x0 = s0 + s2;
1.904 + x1 = s1 + s3;
1.905 + x2 = s0 - s2;
1.906 + x3 = s1 - s3;
1.907 + x4 = fdct_round_shift(s4 + s6);
1.908 + x5 = fdct_round_shift(s5 + s7);
1.909 + x6 = fdct_round_shift(s4 - s6);
1.910 + x7 = fdct_round_shift(s5 - s7);
1.911 + x8 = s8 + s10;
1.912 + x9 = s9 + s11;
1.913 + x10 = s8 - s10;
1.914 + x11 = s9 - s11;
1.915 + x12 = fdct_round_shift(s12 + s14);
1.916 + x13 = fdct_round_shift(s13 + s15);
1.917 + x14 = fdct_round_shift(s12 - s14);
1.918 + x15 = fdct_round_shift(s13 - s15);
1.919 +
1.920 + // stage 4
1.921 + s2 = (- cospi_16_64) * (x2 + x3);
1.922 + s3 = cospi_16_64 * (x2 - x3);
1.923 + s6 = cospi_16_64 * (x6 + x7);
1.924 + s7 = cospi_16_64 * (- x6 + x7);
1.925 + s10 = cospi_16_64 * (x10 + x11);
1.926 + s11 = cospi_16_64 * (- x10 + x11);
1.927 + s14 = (- cospi_16_64) * (x14 + x15);
1.928 + s15 = cospi_16_64 * (x14 - x15);
1.929 +
1.930 + x2 = fdct_round_shift(s2);
1.931 + x3 = fdct_round_shift(s3);
1.932 + x6 = fdct_round_shift(s6);
1.933 + x7 = fdct_round_shift(s7);
1.934 + x10 = fdct_round_shift(s10);
1.935 + x11 = fdct_round_shift(s11);
1.936 + x14 = fdct_round_shift(s14);
1.937 + x15 = fdct_round_shift(s15);
1.938 +
1.939 + output[0] = x0;
1.940 + output[1] = - x8;
1.941 + output[2] = x12;
1.942 + output[3] = - x4;
1.943 + output[4] = x6;
1.944 + output[5] = x14;
1.945 + output[6] = x10;
1.946 + output[7] = x2;
1.947 + output[8] = x3;
1.948 + output[9] = x11;
1.949 + output[10] = x15;
1.950 + output[11] = x7;
1.951 + output[12] = x5;
1.952 + output[13] = - x13;
1.953 + output[14] = x9;
1.954 + output[15] = - x1;
1.955 +}
1.956 +
1.957 +static const transform_2d FHT_16[] = {
1.958 + { fdct16, fdct16 }, // DCT_DCT = 0
1.959 + { fadst16, fdct16 }, // ADST_DCT = 1
1.960 + { fdct16, fadst16 }, // DCT_ADST = 2
1.961 + { fadst16, fadst16 } // ADST_ADST = 3
1.962 +};
1.963 +
1.964 +void vp9_short_fht16x16_c(const int16_t *input, int16_t *output,
1.965 + int stride, int tx_type) {
1.966 + int16_t out[256];
1.967 + int16_t *outptr = &out[0];
1.968 + int i, j;
1.969 + int16_t temp_in[16], temp_out[16];
1.970 + const transform_2d ht = FHT_16[tx_type];
1.971 +
1.972 + // Columns
1.973 + for (i = 0; i < 16; ++i) {
1.974 + for (j = 0; j < 16; ++j)
1.975 + temp_in[j] = input[j * stride + i] * 4;
1.976 + ht.cols(temp_in, temp_out);
1.977 + for (j = 0; j < 16; ++j)
1.978 + outptr[j * 16 + i] = (temp_out[j] + 1 + (temp_out[j] < 0)) >> 2;
1.979 +// outptr[j * 16 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
1.980 + }
1.981 +
1.982 + // Rows
1.983 + for (i = 0; i < 16; ++i) {
1.984 + for (j = 0; j < 16; ++j)
1.985 + temp_in[j] = out[j + i * 16];
1.986 + ht.rows(temp_in, temp_out);
1.987 + for (j = 0; j < 16; ++j)
1.988 + output[j + i * 16] = temp_out[j];
1.989 + }
1.990 +}
1.991 +
1.992 +static INLINE int dct_32_round(int input) {
1.993 + int rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
1.994 + assert(-131072 <= rv && rv <= 131071);
1.995 + return rv;
1.996 +}
1.997 +
1.998 +static INLINE int half_round_shift(int input) {
1.999 + int rv = (input + 1 + (input < 0)) >> 2;
1.1000 + return rv;
1.1001 +}
1.1002 +
1.1003 +static void dct32_1d(const int *input, int *output, int round) {
1.1004 + int step[32];
1.1005 + // Stage 1
1.1006 + step[0] = input[0] + input[(32 - 1)];
1.1007 + step[1] = input[1] + input[(32 - 2)];
1.1008 + step[2] = input[2] + input[(32 - 3)];
1.1009 + step[3] = input[3] + input[(32 - 4)];
1.1010 + step[4] = input[4] + input[(32 - 5)];
1.1011 + step[5] = input[5] + input[(32 - 6)];
1.1012 + step[6] = input[6] + input[(32 - 7)];
1.1013 + step[7] = input[7] + input[(32 - 8)];
1.1014 + step[8] = input[8] + input[(32 - 9)];
1.1015 + step[9] = input[9] + input[(32 - 10)];
1.1016 + step[10] = input[10] + input[(32 - 11)];
1.1017 + step[11] = input[11] + input[(32 - 12)];
1.1018 + step[12] = input[12] + input[(32 - 13)];
1.1019 + step[13] = input[13] + input[(32 - 14)];
1.1020 + step[14] = input[14] + input[(32 - 15)];
1.1021 + step[15] = input[15] + input[(32 - 16)];
1.1022 + step[16] = -input[16] + input[(32 - 17)];
1.1023 + step[17] = -input[17] + input[(32 - 18)];
1.1024 + step[18] = -input[18] + input[(32 - 19)];
1.1025 + step[19] = -input[19] + input[(32 - 20)];
1.1026 + step[20] = -input[20] + input[(32 - 21)];
1.1027 + step[21] = -input[21] + input[(32 - 22)];
1.1028 + step[22] = -input[22] + input[(32 - 23)];
1.1029 + step[23] = -input[23] + input[(32 - 24)];
1.1030 + step[24] = -input[24] + input[(32 - 25)];
1.1031 + step[25] = -input[25] + input[(32 - 26)];
1.1032 + step[26] = -input[26] + input[(32 - 27)];
1.1033 + step[27] = -input[27] + input[(32 - 28)];
1.1034 + step[28] = -input[28] + input[(32 - 29)];
1.1035 + step[29] = -input[29] + input[(32 - 30)];
1.1036 + step[30] = -input[30] + input[(32 - 31)];
1.1037 + step[31] = -input[31] + input[(32 - 32)];
1.1038 +
1.1039 + // Stage 2
1.1040 + output[0] = step[0] + step[16 - 1];
1.1041 + output[1] = step[1] + step[16 - 2];
1.1042 + output[2] = step[2] + step[16 - 3];
1.1043 + output[3] = step[3] + step[16 - 4];
1.1044 + output[4] = step[4] + step[16 - 5];
1.1045 + output[5] = step[5] + step[16 - 6];
1.1046 + output[6] = step[6] + step[16 - 7];
1.1047 + output[7] = step[7] + step[16 - 8];
1.1048 + output[8] = -step[8] + step[16 - 9];
1.1049 + output[9] = -step[9] + step[16 - 10];
1.1050 + output[10] = -step[10] + step[16 - 11];
1.1051 + output[11] = -step[11] + step[16 - 12];
1.1052 + output[12] = -step[12] + step[16 - 13];
1.1053 + output[13] = -step[13] + step[16 - 14];
1.1054 + output[14] = -step[14] + step[16 - 15];
1.1055 + output[15] = -step[15] + step[16 - 16];
1.1056 +
1.1057 + output[16] = step[16];
1.1058 + output[17] = step[17];
1.1059 + output[18] = step[18];
1.1060 + output[19] = step[19];
1.1061 +
1.1062 + output[20] = dct_32_round((-step[20] + step[27]) * cospi_16_64);
1.1063 + output[21] = dct_32_round((-step[21] + step[26]) * cospi_16_64);
1.1064 + output[22] = dct_32_round((-step[22] + step[25]) * cospi_16_64);
1.1065 + output[23] = dct_32_round((-step[23] + step[24]) * cospi_16_64);
1.1066 +
1.1067 + output[24] = dct_32_round((step[24] + step[23]) * cospi_16_64);
1.1068 + output[25] = dct_32_round((step[25] + step[22]) * cospi_16_64);
1.1069 + output[26] = dct_32_round((step[26] + step[21]) * cospi_16_64);
1.1070 + output[27] = dct_32_round((step[27] + step[20]) * cospi_16_64);
1.1071 +
1.1072 + output[28] = step[28];
1.1073 + output[29] = step[29];
1.1074 + output[30] = step[30];
1.1075 + output[31] = step[31];
1.1076 +
1.1077 + // dump the magnitude by 4, hence the intermediate values are within
1.1078 + // the range of 16 bits.
1.1079 + if (round) {
1.1080 + output[0] = half_round_shift(output[0]);
1.1081 + output[1] = half_round_shift(output[1]);
1.1082 + output[2] = half_round_shift(output[2]);
1.1083 + output[3] = half_round_shift(output[3]);
1.1084 + output[4] = half_round_shift(output[4]);
1.1085 + output[5] = half_round_shift(output[5]);
1.1086 + output[6] = half_round_shift(output[6]);
1.1087 + output[7] = half_round_shift(output[7]);
1.1088 + output[8] = half_round_shift(output[8]);
1.1089 + output[9] = half_round_shift(output[9]);
1.1090 + output[10] = half_round_shift(output[10]);
1.1091 + output[11] = half_round_shift(output[11]);
1.1092 + output[12] = half_round_shift(output[12]);
1.1093 + output[13] = half_round_shift(output[13]);
1.1094 + output[14] = half_round_shift(output[14]);
1.1095 + output[15] = half_round_shift(output[15]);
1.1096 +
1.1097 + output[16] = half_round_shift(output[16]);
1.1098 + output[17] = half_round_shift(output[17]);
1.1099 + output[18] = half_round_shift(output[18]);
1.1100 + output[19] = half_round_shift(output[19]);
1.1101 + output[20] = half_round_shift(output[20]);
1.1102 + output[21] = half_round_shift(output[21]);
1.1103 + output[22] = half_round_shift(output[22]);
1.1104 + output[23] = half_round_shift(output[23]);
1.1105 + output[24] = half_round_shift(output[24]);
1.1106 + output[25] = half_round_shift(output[25]);
1.1107 + output[26] = half_round_shift(output[26]);
1.1108 + output[27] = half_round_shift(output[27]);
1.1109 + output[28] = half_round_shift(output[28]);
1.1110 + output[29] = half_round_shift(output[29]);
1.1111 + output[30] = half_round_shift(output[30]);
1.1112 + output[31] = half_round_shift(output[31]);
1.1113 + }
1.1114 +
1.1115 + // Stage 3
1.1116 + step[0] = output[0] + output[(8 - 1)];
1.1117 + step[1] = output[1] + output[(8 - 2)];
1.1118 + step[2] = output[2] + output[(8 - 3)];
1.1119 + step[3] = output[3] + output[(8 - 4)];
1.1120 + step[4] = -output[4] + output[(8 - 5)];
1.1121 + step[5] = -output[5] + output[(8 - 6)];
1.1122 + step[6] = -output[6] + output[(8 - 7)];
1.1123 + step[7] = -output[7] + output[(8 - 8)];
1.1124 + step[8] = output[8];
1.1125 + step[9] = output[9];
1.1126 + step[10] = dct_32_round((-output[10] + output[13]) * cospi_16_64);
1.1127 + step[11] = dct_32_round((-output[11] + output[12]) * cospi_16_64);
1.1128 + step[12] = dct_32_round((output[12] + output[11]) * cospi_16_64);
1.1129 + step[13] = dct_32_round((output[13] + output[10]) * cospi_16_64);
1.1130 + step[14] = output[14];
1.1131 + step[15] = output[15];
1.1132 +
1.1133 + step[16] = output[16] + output[23];
1.1134 + step[17] = output[17] + output[22];
1.1135 + step[18] = output[18] + output[21];
1.1136 + step[19] = output[19] + output[20];
1.1137 + step[20] = -output[20] + output[19];
1.1138 + step[21] = -output[21] + output[18];
1.1139 + step[22] = -output[22] + output[17];
1.1140 + step[23] = -output[23] + output[16];
1.1141 + step[24] = -output[24] + output[31];
1.1142 + step[25] = -output[25] + output[30];
1.1143 + step[26] = -output[26] + output[29];
1.1144 + step[27] = -output[27] + output[28];
1.1145 + step[28] = output[28] + output[27];
1.1146 + step[29] = output[29] + output[26];
1.1147 + step[30] = output[30] + output[25];
1.1148 + step[31] = output[31] + output[24];
1.1149 +
1.1150 + // Stage 4
1.1151 + output[0] = step[0] + step[3];
1.1152 + output[1] = step[1] + step[2];
1.1153 + output[2] = -step[2] + step[1];
1.1154 + output[3] = -step[3] + step[0];
1.1155 + output[4] = step[4];
1.1156 + output[5] = dct_32_round((-step[5] + step[6]) * cospi_16_64);
1.1157 + output[6] = dct_32_round((step[6] + step[5]) * cospi_16_64);
1.1158 + output[7] = step[7];
1.1159 + output[8] = step[8] + step[11];
1.1160 + output[9] = step[9] + step[10];
1.1161 + output[10] = -step[10] + step[9];
1.1162 + output[11] = -step[11] + step[8];
1.1163 + output[12] = -step[12] + step[15];
1.1164 + output[13] = -step[13] + step[14];
1.1165 + output[14] = step[14] + step[13];
1.1166 + output[15] = step[15] + step[12];
1.1167 +
1.1168 + output[16] = step[16];
1.1169 + output[17] = step[17];
1.1170 + output[18] = dct_32_round(step[18] * -cospi_8_64 + step[29] * cospi_24_64);
1.1171 + output[19] = dct_32_round(step[19] * -cospi_8_64 + step[28] * cospi_24_64);
1.1172 + output[20] = dct_32_round(step[20] * -cospi_24_64 + step[27] * -cospi_8_64);
1.1173 + output[21] = dct_32_round(step[21] * -cospi_24_64 + step[26] * -cospi_8_64);
1.1174 + output[22] = step[22];
1.1175 + output[23] = step[23];
1.1176 + output[24] = step[24];
1.1177 + output[25] = step[25];
1.1178 + output[26] = dct_32_round(step[26] * cospi_24_64 + step[21] * -cospi_8_64);
1.1179 + output[27] = dct_32_round(step[27] * cospi_24_64 + step[20] * -cospi_8_64);
1.1180 + output[28] = dct_32_round(step[28] * cospi_8_64 + step[19] * cospi_24_64);
1.1181 + output[29] = dct_32_round(step[29] * cospi_8_64 + step[18] * cospi_24_64);
1.1182 + output[30] = step[30];
1.1183 + output[31] = step[31];
1.1184 +
1.1185 + // Stage 5
1.1186 + step[0] = dct_32_round((output[0] + output[1]) * cospi_16_64);
1.1187 + step[1] = dct_32_round((-output[1] + output[0]) * cospi_16_64);
1.1188 + step[2] = dct_32_round(output[2] * cospi_24_64 + output[3] * cospi_8_64);
1.1189 + step[3] = dct_32_round(output[3] * cospi_24_64 - output[2] * cospi_8_64);
1.1190 + step[4] = output[4] + output[5];
1.1191 + step[5] = -output[5] + output[4];
1.1192 + step[6] = -output[6] + output[7];
1.1193 + step[7] = output[7] + output[6];
1.1194 + step[8] = output[8];
1.1195 + step[9] = dct_32_round(output[9] * -cospi_8_64 + output[14] * cospi_24_64);
1.1196 + step[10] = dct_32_round(output[10] * -cospi_24_64 + output[13] * -cospi_8_64);
1.1197 + step[11] = output[11];
1.1198 + step[12] = output[12];
1.1199 + step[13] = dct_32_round(output[13] * cospi_24_64 + output[10] * -cospi_8_64);
1.1200 + step[14] = dct_32_round(output[14] * cospi_8_64 + output[9] * cospi_24_64);
1.1201 + step[15] = output[15];
1.1202 +
1.1203 + step[16] = output[16] + output[19];
1.1204 + step[17] = output[17] + output[18];
1.1205 + step[18] = -output[18] + output[17];
1.1206 + step[19] = -output[19] + output[16];
1.1207 + step[20] = -output[20] + output[23];
1.1208 + step[21] = -output[21] + output[22];
1.1209 + step[22] = output[22] + output[21];
1.1210 + step[23] = output[23] + output[20];
1.1211 + step[24] = output[24] + output[27];
1.1212 + step[25] = output[25] + output[26];
1.1213 + step[26] = -output[26] + output[25];
1.1214 + step[27] = -output[27] + output[24];
1.1215 + step[28] = -output[28] + output[31];
1.1216 + step[29] = -output[29] + output[30];
1.1217 + step[30] = output[30] + output[29];
1.1218 + step[31] = output[31] + output[28];
1.1219 +
1.1220 + // Stage 6
1.1221 + output[0] = step[0];
1.1222 + output[1] = step[1];
1.1223 + output[2] = step[2];
1.1224 + output[3] = step[3];
1.1225 + output[4] = dct_32_round(step[4] * cospi_28_64 + step[7] * cospi_4_64);
1.1226 + output[5] = dct_32_round(step[5] * cospi_12_64 + step[6] * cospi_20_64);
1.1227 + output[6] = dct_32_round(step[6] * cospi_12_64 + step[5] * -cospi_20_64);
1.1228 + output[7] = dct_32_round(step[7] * cospi_28_64 + step[4] * -cospi_4_64);
1.1229 + output[8] = step[8] + step[9];
1.1230 + output[9] = -step[9] + step[8];
1.1231 + output[10] = -step[10] + step[11];
1.1232 + output[11] = step[11] + step[10];
1.1233 + output[12] = step[12] + step[13];
1.1234 + output[13] = -step[13] + step[12];
1.1235 + output[14] = -step[14] + step[15];
1.1236 + output[15] = step[15] + step[14];
1.1237 +
1.1238 + output[16] = step[16];
1.1239 + output[17] = dct_32_round(step[17] * -cospi_4_64 + step[30] * cospi_28_64);
1.1240 + output[18] = dct_32_round(step[18] * -cospi_28_64 + step[29] * -cospi_4_64);
1.1241 + output[19] = step[19];
1.1242 + output[20] = step[20];
1.1243 + output[21] = dct_32_round(step[21] * -cospi_20_64 + step[26] * cospi_12_64);
1.1244 + output[22] = dct_32_round(step[22] * -cospi_12_64 + step[25] * -cospi_20_64);
1.1245 + output[23] = step[23];
1.1246 + output[24] = step[24];
1.1247 + output[25] = dct_32_round(step[25] * cospi_12_64 + step[22] * -cospi_20_64);
1.1248 + output[26] = dct_32_round(step[26] * cospi_20_64 + step[21] * cospi_12_64);
1.1249 + output[27] = step[27];
1.1250 + output[28] = step[28];
1.1251 + output[29] = dct_32_round(step[29] * cospi_28_64 + step[18] * -cospi_4_64);
1.1252 + output[30] = dct_32_round(step[30] * cospi_4_64 + step[17] * cospi_28_64);
1.1253 + output[31] = step[31];
1.1254 +
1.1255 + // Stage 7
1.1256 + step[0] = output[0];
1.1257 + step[1] = output[1];
1.1258 + step[2] = output[2];
1.1259 + step[3] = output[3];
1.1260 + step[4] = output[4];
1.1261 + step[5] = output[5];
1.1262 + step[6] = output[6];
1.1263 + step[7] = output[7];
1.1264 + step[8] = dct_32_round(output[8] * cospi_30_64 + output[15] * cospi_2_64);
1.1265 + step[9] = dct_32_round(output[9] * cospi_14_64 + output[14] * cospi_18_64);
1.1266 + step[10] = dct_32_round(output[10] * cospi_22_64 + output[13] * cospi_10_64);
1.1267 + step[11] = dct_32_round(output[11] * cospi_6_64 + output[12] * cospi_26_64);
1.1268 + step[12] = dct_32_round(output[12] * cospi_6_64 + output[11] * -cospi_26_64);
1.1269 + step[13] = dct_32_round(output[13] * cospi_22_64 + output[10] * -cospi_10_64);
1.1270 + step[14] = dct_32_round(output[14] * cospi_14_64 + output[9] * -cospi_18_64);
1.1271 + step[15] = dct_32_round(output[15] * cospi_30_64 + output[8] * -cospi_2_64);
1.1272 +
1.1273 + step[16] = output[16] + output[17];
1.1274 + step[17] = -output[17] + output[16];
1.1275 + step[18] = -output[18] + output[19];
1.1276 + step[19] = output[19] + output[18];
1.1277 + step[20] = output[20] + output[21];
1.1278 + step[21] = -output[21] + output[20];
1.1279 + step[22] = -output[22] + output[23];
1.1280 + step[23] = output[23] + output[22];
1.1281 + step[24] = output[24] + output[25];
1.1282 + step[25] = -output[25] + output[24];
1.1283 + step[26] = -output[26] + output[27];
1.1284 + step[27] = output[27] + output[26];
1.1285 + step[28] = output[28] + output[29];
1.1286 + step[29] = -output[29] + output[28];
1.1287 + step[30] = -output[30] + output[31];
1.1288 + step[31] = output[31] + output[30];
1.1289 +
1.1290 + // Final stage --- outputs indices are bit-reversed.
1.1291 + output[0] = step[0];
1.1292 + output[16] = step[1];
1.1293 + output[8] = step[2];
1.1294 + output[24] = step[3];
1.1295 + output[4] = step[4];
1.1296 + output[20] = step[5];
1.1297 + output[12] = step[6];
1.1298 + output[28] = step[7];
1.1299 + output[2] = step[8];
1.1300 + output[18] = step[9];
1.1301 + output[10] = step[10];
1.1302 + output[26] = step[11];
1.1303 + output[6] = step[12];
1.1304 + output[22] = step[13];
1.1305 + output[14] = step[14];
1.1306 + output[30] = step[15];
1.1307 +
1.1308 + output[1] = dct_32_round(step[16] * cospi_31_64 + step[31] * cospi_1_64);
1.1309 + output[17] = dct_32_round(step[17] * cospi_15_64 + step[30] * cospi_17_64);
1.1310 + output[9] = dct_32_round(step[18] * cospi_23_64 + step[29] * cospi_9_64);
1.1311 + output[25] = dct_32_round(step[19] * cospi_7_64 + step[28] * cospi_25_64);
1.1312 + output[5] = dct_32_round(step[20] * cospi_27_64 + step[27] * cospi_5_64);
1.1313 + output[21] = dct_32_round(step[21] * cospi_11_64 + step[26] * cospi_21_64);
1.1314 + output[13] = dct_32_round(step[22] * cospi_19_64 + step[25] * cospi_13_64);
1.1315 + output[29] = dct_32_round(step[23] * cospi_3_64 + step[24] * cospi_29_64);
1.1316 + output[3] = dct_32_round(step[24] * cospi_3_64 + step[23] * -cospi_29_64);
1.1317 + output[19] = dct_32_round(step[25] * cospi_19_64 + step[22] * -cospi_13_64);
1.1318 + output[11] = dct_32_round(step[26] * cospi_11_64 + step[21] * -cospi_21_64);
1.1319 + output[27] = dct_32_round(step[27] * cospi_27_64 + step[20] * -cospi_5_64);
1.1320 + output[7] = dct_32_round(step[28] * cospi_7_64 + step[19] * -cospi_25_64);
1.1321 + output[23] = dct_32_round(step[29] * cospi_23_64 + step[18] * -cospi_9_64);
1.1322 + output[15] = dct_32_round(step[30] * cospi_15_64 + step[17] * -cospi_17_64);
1.1323 + output[31] = dct_32_round(step[31] * cospi_31_64 + step[16] * -cospi_1_64);
1.1324 +}
1.1325 +
1.1326 +void vp9_fdct32x32_c(const int16_t *input, int16_t *out, int stride) {
1.1327 + int i, j;
1.1328 + int output[32 * 32];
1.1329 +
1.1330 + // Columns
1.1331 + for (i = 0; i < 32; ++i) {
1.1332 + int temp_in[32], temp_out[32];
1.1333 + for (j = 0; j < 32; ++j)
1.1334 + temp_in[j] = input[j * stride + i] * 4;
1.1335 + dct32_1d(temp_in, temp_out, 0);
1.1336 + for (j = 0; j < 32; ++j)
1.1337 + output[j * 32 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
1.1338 + }
1.1339 +
1.1340 + // Rows
1.1341 + for (i = 0; i < 32; ++i) {
1.1342 + int temp_in[32], temp_out[32];
1.1343 + for (j = 0; j < 32; ++j)
1.1344 + temp_in[j] = output[j + i * 32];
1.1345 + dct32_1d(temp_in, temp_out, 0);
1.1346 + for (j = 0; j < 32; ++j)
1.1347 + out[j + i * 32] = (temp_out[j] + 1 + (temp_out[j] < 0)) >> 2;
1.1348 + }
1.1349 +}
1.1350 +
1.1351 +// Note that although we use dct_32_round in dct32_1d computation flow,
1.1352 +// this 2d fdct32x32 for rate-distortion optimization loop is operating
1.1353 +// within 16 bits precision.
1.1354 +void vp9_fdct32x32_rd_c(const int16_t *input, int16_t *out, int stride) {
1.1355 + int i, j;
1.1356 + int output[32 * 32];
1.1357 +
1.1358 + // Columns
1.1359 + for (i = 0; i < 32; ++i) {
1.1360 + int temp_in[32], temp_out[32];
1.1361 + for (j = 0; j < 32; ++j)
1.1362 + temp_in[j] = input[j * stride + i] * 4;
1.1363 + dct32_1d(temp_in, temp_out, 0);
1.1364 + for (j = 0; j < 32; ++j)
1.1365 + // TODO(cd): see quality impact of only doing
1.1366 + // output[j * 32 + i] = (temp_out[j] + 1) >> 2;
1.1367 + // PS: also change code in vp9/encoder/x86/vp9_dct_sse2.c
1.1368 + output[j * 32 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
1.1369 + }
1.1370 +
1.1371 + // Rows
1.1372 + for (i = 0; i < 32; ++i) {
1.1373 + int temp_in[32], temp_out[32];
1.1374 + for (j = 0; j < 32; ++j)
1.1375 + temp_in[j] = output[j + i * 32];
1.1376 + dct32_1d(temp_in, temp_out, 1);
1.1377 + for (j = 0; j < 32; ++j)
1.1378 + out[j + i * 32] = temp_out[j];
1.1379 + }
1.1380 +}
1.1381 +
1.1382 +void vp9_fht4x4(TX_TYPE tx_type, const int16_t *input, int16_t *output,
1.1383 + int stride) {
1.1384 + if (tx_type == DCT_DCT)
1.1385 + vp9_fdct4x4(input, output, stride);
1.1386 + else
1.1387 + vp9_short_fht4x4(input, output, stride, tx_type);
1.1388 +}
1.1389 +
1.1390 +void vp9_fht8x8(TX_TYPE tx_type, const int16_t *input, int16_t *output,
1.1391 + int stride) {
1.1392 + if (tx_type == DCT_DCT)
1.1393 + vp9_fdct8x8(input, output, stride);
1.1394 + else
1.1395 + vp9_short_fht8x8(input, output, stride, tx_type);
1.1396 +}
1.1397 +
1.1398 +void vp9_fht16x16(TX_TYPE tx_type, const int16_t *input, int16_t *output,
1.1399 + int stride) {
1.1400 + if (tx_type == DCT_DCT)
1.1401 + vp9_fdct16x16(input, output, stride);
1.1402 + else
1.1403 + vp9_short_fht16x16(input, output, stride, tx_type);
1.1404 +}
