1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp9/common/vp9_idct.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,1416 @@
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 +#include "vp9/common/vp9_systemdependent.h"
1.20 +#include "vp9/common/vp9_blockd.h"
1.21 +#include "vp9/common/vp9_common.h"
1.22 +#include "vp9/common/vp9_idct.h"
1.23 +
1.24 +void vp9_iwht4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride) {
1.25 +/* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
1.26 + 0.5 shifts per pixel. */
1.27 + int i;
1.28 + int16_t output[16];
1.29 + int a1, b1, c1, d1, e1;
1.30 + const int16_t *ip = input;
1.31 + int16_t *op = output;
1.32 +
1.33 + for (i = 0; i < 4; i++) {
1.34 + a1 = ip[0] >> UNIT_QUANT_SHIFT;
1.35 + c1 = ip[1] >> UNIT_QUANT_SHIFT;
1.36 + d1 = ip[2] >> UNIT_QUANT_SHIFT;
1.37 + b1 = ip[3] >> UNIT_QUANT_SHIFT;
1.38 + a1 += c1;
1.39 + d1 -= b1;
1.40 + e1 = (a1 - d1) >> 1;
1.41 + b1 = e1 - b1;
1.42 + c1 = e1 - c1;
1.43 + a1 -= b1;
1.44 + d1 += c1;
1.45 + op[0] = a1;
1.46 + op[1] = b1;
1.47 + op[2] = c1;
1.48 + op[3] = d1;
1.49 + ip += 4;
1.50 + op += 4;
1.51 + }
1.52 +
1.53 + ip = output;
1.54 + for (i = 0; i < 4; i++) {
1.55 + a1 = ip[4 * 0];
1.56 + c1 = ip[4 * 1];
1.57 + d1 = ip[4 * 2];
1.58 + b1 = ip[4 * 3];
1.59 + a1 += c1;
1.60 + d1 -= b1;
1.61 + e1 = (a1 - d1) >> 1;
1.62 + b1 = e1 - b1;
1.63 + c1 = e1 - c1;
1.64 + a1 -= b1;
1.65 + d1 += c1;
1.66 + dest[stride * 0] = clip_pixel(dest[stride * 0] + a1);
1.67 + dest[stride * 1] = clip_pixel(dest[stride * 1] + b1);
1.68 + dest[stride * 2] = clip_pixel(dest[stride * 2] + c1);
1.69 + dest[stride * 3] = clip_pixel(dest[stride * 3] + d1);
1.70 +
1.71 + ip++;
1.72 + dest++;
1.73 + }
1.74 +}
1.75 +
1.76 +void vp9_iwht4x4_1_add_c(const int16_t *in, uint8_t *dest, int dest_stride) {
1.77 + int i;
1.78 + int a1, e1;
1.79 + int16_t tmp[4];
1.80 + const int16_t *ip = in;
1.81 + int16_t *op = tmp;
1.82 +
1.83 + a1 = ip[0] >> UNIT_QUANT_SHIFT;
1.84 + e1 = a1 >> 1;
1.85 + a1 -= e1;
1.86 + op[0] = a1;
1.87 + op[1] = op[2] = op[3] = e1;
1.88 +
1.89 + ip = tmp;
1.90 + for (i = 0; i < 4; i++) {
1.91 + e1 = ip[0] >> 1;
1.92 + a1 = ip[0] - e1;
1.93 + dest[dest_stride * 0] = clip_pixel(dest[dest_stride * 0] + a1);
1.94 + dest[dest_stride * 1] = clip_pixel(dest[dest_stride * 1] + e1);
1.95 + dest[dest_stride * 2] = clip_pixel(dest[dest_stride * 2] + e1);
1.96 + dest[dest_stride * 3] = clip_pixel(dest[dest_stride * 3] + e1);
1.97 + ip++;
1.98 + dest++;
1.99 + }
1.100 +}
1.101 +
1.102 +static void idct4_1d(const int16_t *input, int16_t *output) {
1.103 + int16_t step[4];
1.104 + int temp1, temp2;
1.105 + // stage 1
1.106 + temp1 = (input[0] + input[2]) * cospi_16_64;
1.107 + temp2 = (input[0] - input[2]) * cospi_16_64;
1.108 + step[0] = dct_const_round_shift(temp1);
1.109 + step[1] = dct_const_round_shift(temp2);
1.110 + temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64;
1.111 + temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64;
1.112 + step[2] = dct_const_round_shift(temp1);
1.113 + step[3] = dct_const_round_shift(temp2);
1.114 +
1.115 + // stage 2
1.116 + output[0] = step[0] + step[3];
1.117 + output[1] = step[1] + step[2];
1.118 + output[2] = step[1] - step[2];
1.119 + output[3] = step[0] - step[3];
1.120 +}
1.121 +
1.122 +void vp9_idct4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride) {
1.123 + int16_t out[4 * 4];
1.124 + int16_t *outptr = out;
1.125 + int i, j;
1.126 + int16_t temp_in[4], temp_out[4];
1.127 +
1.128 + // Rows
1.129 + for (i = 0; i < 4; ++i) {
1.130 + idct4_1d(input, outptr);
1.131 + input += 4;
1.132 + outptr += 4;
1.133 + }
1.134 +
1.135 + // Columns
1.136 + for (i = 0; i < 4; ++i) {
1.137 + for (j = 0; j < 4; ++j)
1.138 + temp_in[j] = out[j * 4 + i];
1.139 + idct4_1d(temp_in, temp_out);
1.140 + for (j = 0; j < 4; ++j)
1.141 + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
1.142 + + dest[j * stride + i]);
1.143 + }
1.144 +}
1.145 +
1.146 +void vp9_idct4x4_1_add_c(const int16_t *input, uint8_t *dest, int dest_stride) {
1.147 + int i;
1.148 + int a1;
1.149 + int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
1.150 + out = dct_const_round_shift(out * cospi_16_64);
1.151 + a1 = ROUND_POWER_OF_TWO(out, 4);
1.152 +
1.153 + for (i = 0; i < 4; i++) {
1.154 + dest[0] = clip_pixel(dest[0] + a1);
1.155 + dest[1] = clip_pixel(dest[1] + a1);
1.156 + dest[2] = clip_pixel(dest[2] + a1);
1.157 + dest[3] = clip_pixel(dest[3] + a1);
1.158 + dest += dest_stride;
1.159 + }
1.160 +}
1.161 +
1.162 +static void idct8_1d(const int16_t *input, int16_t *output) {
1.163 + int16_t step1[8], step2[8];
1.164 + int temp1, temp2;
1.165 + // stage 1
1.166 + step1[0] = input[0];
1.167 + step1[2] = input[4];
1.168 + step1[1] = input[2];
1.169 + step1[3] = input[6];
1.170 + temp1 = input[1] * cospi_28_64 - input[7] * cospi_4_64;
1.171 + temp2 = input[1] * cospi_4_64 + input[7] * cospi_28_64;
1.172 + step1[4] = dct_const_round_shift(temp1);
1.173 + step1[7] = dct_const_round_shift(temp2);
1.174 + temp1 = input[5] * cospi_12_64 - input[3] * cospi_20_64;
1.175 + temp2 = input[5] * cospi_20_64 + input[3] * cospi_12_64;
1.176 + step1[5] = dct_const_round_shift(temp1);
1.177 + step1[6] = dct_const_round_shift(temp2);
1.178 +
1.179 + // stage 2 & stage 3 - even half
1.180 + idct4_1d(step1, step1);
1.181 +
1.182 + // stage 2 - odd half
1.183 + step2[4] = step1[4] + step1[5];
1.184 + step2[5] = step1[4] - step1[5];
1.185 + step2[6] = -step1[6] + step1[7];
1.186 + step2[7] = step1[6] + step1[7];
1.187 +
1.188 + // stage 3 -odd half
1.189 + step1[4] = step2[4];
1.190 + temp1 = (step2[6] - step2[5]) * cospi_16_64;
1.191 + temp2 = (step2[5] + step2[6]) * cospi_16_64;
1.192 + step1[5] = dct_const_round_shift(temp1);
1.193 + step1[6] = dct_const_round_shift(temp2);
1.194 + step1[7] = step2[7];
1.195 +
1.196 + // stage 4
1.197 + output[0] = step1[0] + step1[7];
1.198 + output[1] = step1[1] + step1[6];
1.199 + output[2] = step1[2] + step1[5];
1.200 + output[3] = step1[3] + step1[4];
1.201 + output[4] = step1[3] - step1[4];
1.202 + output[5] = step1[2] - step1[5];
1.203 + output[6] = step1[1] - step1[6];
1.204 + output[7] = step1[0] - step1[7];
1.205 +}
1.206 +
1.207 +void vp9_idct8x8_64_add_c(const int16_t *input, uint8_t *dest, int stride) {
1.208 + int16_t out[8 * 8];
1.209 + int16_t *outptr = out;
1.210 + int i, j;
1.211 + int16_t temp_in[8], temp_out[8];
1.212 +
1.213 + // First transform rows
1.214 + for (i = 0; i < 8; ++i) {
1.215 + idct8_1d(input, outptr);
1.216 + input += 8;
1.217 + outptr += 8;
1.218 + }
1.219 +
1.220 + // Then transform columns
1.221 + for (i = 0; i < 8; ++i) {
1.222 + for (j = 0; j < 8; ++j)
1.223 + temp_in[j] = out[j * 8 + i];
1.224 + idct8_1d(temp_in, temp_out);
1.225 + for (j = 0; j < 8; ++j)
1.226 + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
1.227 + + dest[j * stride + i]);
1.228 + }
1.229 +}
1.230 +
1.231 +void vp9_idct8x8_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
1.232 + int i, j;
1.233 + int a1;
1.234 + int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
1.235 + out = dct_const_round_shift(out * cospi_16_64);
1.236 + a1 = ROUND_POWER_OF_TWO(out, 5);
1.237 + for (j = 0; j < 8; ++j) {
1.238 + for (i = 0; i < 8; ++i)
1.239 + dest[i] = clip_pixel(dest[i] + a1);
1.240 + dest += stride;
1.241 + }
1.242 +}
1.243 +
1.244 +static void iadst4_1d(const int16_t *input, int16_t *output) {
1.245 + int s0, s1, s2, s3, s4, s5, s6, s7;
1.246 +
1.247 + int x0 = input[0];
1.248 + int x1 = input[1];
1.249 + int x2 = input[2];
1.250 + int x3 = input[3];
1.251 +
1.252 + if (!(x0 | x1 | x2 | x3)) {
1.253 + output[0] = output[1] = output[2] = output[3] = 0;
1.254 + return;
1.255 + }
1.256 +
1.257 + s0 = sinpi_1_9 * x0;
1.258 + s1 = sinpi_2_9 * x0;
1.259 + s2 = sinpi_3_9 * x1;
1.260 + s3 = sinpi_4_9 * x2;
1.261 + s4 = sinpi_1_9 * x2;
1.262 + s5 = sinpi_2_9 * x3;
1.263 + s6 = sinpi_4_9 * x3;
1.264 + s7 = x0 - x2 + x3;
1.265 +
1.266 + x0 = s0 + s3 + s5;
1.267 + x1 = s1 - s4 - s6;
1.268 + x2 = sinpi_3_9 * s7;
1.269 + x3 = s2;
1.270 +
1.271 + s0 = x0 + x3;
1.272 + s1 = x1 + x3;
1.273 + s2 = x2;
1.274 + s3 = x0 + x1 - x3;
1.275 +
1.276 + // 1-D transform scaling factor is sqrt(2).
1.277 + // The overall dynamic range is 14b (input) + 14b (multiplication scaling)
1.278 + // + 1b (addition) = 29b.
1.279 + // Hence the output bit depth is 15b.
1.280 + output[0] = dct_const_round_shift(s0);
1.281 + output[1] = dct_const_round_shift(s1);
1.282 + output[2] = dct_const_round_shift(s2);
1.283 + output[3] = dct_const_round_shift(s3);
1.284 +}
1.285 +
1.286 +void vp9_iht4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride,
1.287 + int tx_type) {
1.288 + const transform_2d IHT_4[] = {
1.289 + { idct4_1d, idct4_1d }, // DCT_DCT = 0
1.290 + { iadst4_1d, idct4_1d }, // ADST_DCT = 1
1.291 + { idct4_1d, iadst4_1d }, // DCT_ADST = 2
1.292 + { iadst4_1d, iadst4_1d } // ADST_ADST = 3
1.293 + };
1.294 +
1.295 + int i, j;
1.296 + int16_t out[4 * 4];
1.297 + int16_t *outptr = out;
1.298 + int16_t temp_in[4], temp_out[4];
1.299 +
1.300 + // inverse transform row vectors
1.301 + for (i = 0; i < 4; ++i) {
1.302 + IHT_4[tx_type].rows(input, outptr);
1.303 + input += 4;
1.304 + outptr += 4;
1.305 + }
1.306 +
1.307 + // inverse transform column vectors
1.308 + for (i = 0; i < 4; ++i) {
1.309 + for (j = 0; j < 4; ++j)
1.310 + temp_in[j] = out[j * 4 + i];
1.311 + IHT_4[tx_type].cols(temp_in, temp_out);
1.312 + for (j = 0; j < 4; ++j)
1.313 + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
1.314 + + dest[j * stride + i]);
1.315 + }
1.316 +}
1.317 +static void iadst8_1d(const int16_t *input, int16_t *output) {
1.318 + int s0, s1, s2, s3, s4, s5, s6, s7;
1.319 +
1.320 + int x0 = input[7];
1.321 + int x1 = input[0];
1.322 + int x2 = input[5];
1.323 + int x3 = input[2];
1.324 + int x4 = input[3];
1.325 + int x5 = input[4];
1.326 + int x6 = input[1];
1.327 + int x7 = input[6];
1.328 +
1.329 + if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7)) {
1.330 + output[0] = output[1] = output[2] = output[3] = output[4]
1.331 + = output[5] = output[6] = output[7] = 0;
1.332 + return;
1.333 + }
1.334 +
1.335 + // stage 1
1.336 + s0 = cospi_2_64 * x0 + cospi_30_64 * x1;
1.337 + s1 = cospi_30_64 * x0 - cospi_2_64 * x1;
1.338 + s2 = cospi_10_64 * x2 + cospi_22_64 * x3;
1.339 + s3 = cospi_22_64 * x2 - cospi_10_64 * x3;
1.340 + s4 = cospi_18_64 * x4 + cospi_14_64 * x5;
1.341 + s5 = cospi_14_64 * x4 - cospi_18_64 * x5;
1.342 + s6 = cospi_26_64 * x6 + cospi_6_64 * x7;
1.343 + s7 = cospi_6_64 * x6 - cospi_26_64 * x7;
1.344 +
1.345 + x0 = dct_const_round_shift(s0 + s4);
1.346 + x1 = dct_const_round_shift(s1 + s5);
1.347 + x2 = dct_const_round_shift(s2 + s6);
1.348 + x3 = dct_const_round_shift(s3 + s7);
1.349 + x4 = dct_const_round_shift(s0 - s4);
1.350 + x5 = dct_const_round_shift(s1 - s5);
1.351 + x6 = dct_const_round_shift(s2 - s6);
1.352 + x7 = dct_const_round_shift(s3 - s7);
1.353 +
1.354 + // stage 2
1.355 + s0 = x0;
1.356 + s1 = x1;
1.357 + s2 = x2;
1.358 + s3 = x3;
1.359 + s4 = cospi_8_64 * x4 + cospi_24_64 * x5;
1.360 + s5 = cospi_24_64 * x4 - cospi_8_64 * x5;
1.361 + s6 = -cospi_24_64 * x6 + cospi_8_64 * x7;
1.362 + s7 = cospi_8_64 * x6 + cospi_24_64 * x7;
1.363 +
1.364 + x0 = s0 + s2;
1.365 + x1 = s1 + s3;
1.366 + x2 = s0 - s2;
1.367 + x3 = s1 - s3;
1.368 + x4 = dct_const_round_shift(s4 + s6);
1.369 + x5 = dct_const_round_shift(s5 + s7);
1.370 + x6 = dct_const_round_shift(s4 - s6);
1.371 + x7 = dct_const_round_shift(s5 - s7);
1.372 +
1.373 + // stage 3
1.374 + s2 = cospi_16_64 * (x2 + x3);
1.375 + s3 = cospi_16_64 * (x2 - x3);
1.376 + s6 = cospi_16_64 * (x6 + x7);
1.377 + s7 = cospi_16_64 * (x6 - x7);
1.378 +
1.379 + x2 = dct_const_round_shift(s2);
1.380 + x3 = dct_const_round_shift(s3);
1.381 + x6 = dct_const_round_shift(s6);
1.382 + x7 = dct_const_round_shift(s7);
1.383 +
1.384 + output[0] = x0;
1.385 + output[1] = -x4;
1.386 + output[2] = x6;
1.387 + output[3] = -x2;
1.388 + output[4] = x3;
1.389 + output[5] = -x7;
1.390 + output[6] = x5;
1.391 + output[7] = -x1;
1.392 +}
1.393 +
1.394 +static const transform_2d IHT_8[] = {
1.395 + { idct8_1d, idct8_1d }, // DCT_DCT = 0
1.396 + { iadst8_1d, idct8_1d }, // ADST_DCT = 1
1.397 + { idct8_1d, iadst8_1d }, // DCT_ADST = 2
1.398 + { iadst8_1d, iadst8_1d } // ADST_ADST = 3
1.399 +};
1.400 +
1.401 +void vp9_iht8x8_64_add_c(const int16_t *input, uint8_t *dest, int stride,
1.402 + int tx_type) {
1.403 + int i, j;
1.404 + int16_t out[8 * 8];
1.405 + int16_t *outptr = out;
1.406 + int16_t temp_in[8], temp_out[8];
1.407 + const transform_2d ht = IHT_8[tx_type];
1.408 +
1.409 + // inverse transform row vectors
1.410 + for (i = 0; i < 8; ++i) {
1.411 + ht.rows(input, outptr);
1.412 + input += 8;
1.413 + outptr += 8;
1.414 + }
1.415 +
1.416 + // inverse transform column vectors
1.417 + for (i = 0; i < 8; ++i) {
1.418 + for (j = 0; j < 8; ++j)
1.419 + temp_in[j] = out[j * 8 + i];
1.420 + ht.cols(temp_in, temp_out);
1.421 + for (j = 0; j < 8; ++j)
1.422 + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
1.423 + + dest[j * stride + i]);
1.424 + }
1.425 +}
1.426 +
1.427 +void vp9_idct8x8_10_add_c(const int16_t *input, uint8_t *dest, int stride) {
1.428 + int16_t out[8 * 8] = { 0 };
1.429 + int16_t *outptr = out;
1.430 + int i, j;
1.431 + int16_t temp_in[8], temp_out[8];
1.432 +
1.433 + // First transform rows
1.434 + // only first 4 row has non-zero coefs
1.435 + for (i = 0; i < 4; ++i) {
1.436 + idct8_1d(input, outptr);
1.437 + input += 8;
1.438 + outptr += 8;
1.439 + }
1.440 +
1.441 + // Then transform columns
1.442 + for (i = 0; i < 8; ++i) {
1.443 + for (j = 0; j < 8; ++j)
1.444 + temp_in[j] = out[j * 8 + i];
1.445 + idct8_1d(temp_in, temp_out);
1.446 + for (j = 0; j < 8; ++j)
1.447 + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
1.448 + + dest[j * stride + i]);
1.449 + }
1.450 +}
1.451 +
1.452 +static void idct16_1d(const int16_t *input, int16_t *output) {
1.453 + int16_t step1[16], step2[16];
1.454 + int temp1, temp2;
1.455 +
1.456 + // stage 1
1.457 + step1[0] = input[0/2];
1.458 + step1[1] = input[16/2];
1.459 + step1[2] = input[8/2];
1.460 + step1[3] = input[24/2];
1.461 + step1[4] = input[4/2];
1.462 + step1[5] = input[20/2];
1.463 + step1[6] = input[12/2];
1.464 + step1[7] = input[28/2];
1.465 + step1[8] = input[2/2];
1.466 + step1[9] = input[18/2];
1.467 + step1[10] = input[10/2];
1.468 + step1[11] = input[26/2];
1.469 + step1[12] = input[6/2];
1.470 + step1[13] = input[22/2];
1.471 + step1[14] = input[14/2];
1.472 + step1[15] = input[30/2];
1.473 +
1.474 + // stage 2
1.475 + step2[0] = step1[0];
1.476 + step2[1] = step1[1];
1.477 + step2[2] = step1[2];
1.478 + step2[3] = step1[3];
1.479 + step2[4] = step1[4];
1.480 + step2[5] = step1[5];
1.481 + step2[6] = step1[6];
1.482 + step2[7] = step1[7];
1.483 +
1.484 + temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64;
1.485 + temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64;
1.486 + step2[8] = dct_const_round_shift(temp1);
1.487 + step2[15] = dct_const_round_shift(temp2);
1.488 +
1.489 + temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64;
1.490 + temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64;
1.491 + step2[9] = dct_const_round_shift(temp1);
1.492 + step2[14] = dct_const_round_shift(temp2);
1.493 +
1.494 + temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64;
1.495 + temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64;
1.496 + step2[10] = dct_const_round_shift(temp1);
1.497 + step2[13] = dct_const_round_shift(temp2);
1.498 +
1.499 + temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64;
1.500 + temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64;
1.501 + step2[11] = dct_const_round_shift(temp1);
1.502 + step2[12] = dct_const_round_shift(temp2);
1.503 +
1.504 + // stage 3
1.505 + step1[0] = step2[0];
1.506 + step1[1] = step2[1];
1.507 + step1[2] = step2[2];
1.508 + step1[3] = step2[3];
1.509 +
1.510 + temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64;
1.511 + temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64;
1.512 + step1[4] = dct_const_round_shift(temp1);
1.513 + step1[7] = dct_const_round_shift(temp2);
1.514 + temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64;
1.515 + temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64;
1.516 + step1[5] = dct_const_round_shift(temp1);
1.517 + step1[6] = dct_const_round_shift(temp2);
1.518 +
1.519 + step1[8] = step2[8] + step2[9];
1.520 + step1[9] = step2[8] - step2[9];
1.521 + step1[10] = -step2[10] + step2[11];
1.522 + step1[11] = step2[10] + step2[11];
1.523 + step1[12] = step2[12] + step2[13];
1.524 + step1[13] = step2[12] - step2[13];
1.525 + step1[14] = -step2[14] + step2[15];
1.526 + step1[15] = step2[14] + step2[15];
1.527 +
1.528 + // stage 4
1.529 + temp1 = (step1[0] + step1[1]) * cospi_16_64;
1.530 + temp2 = (step1[0] - step1[1]) * cospi_16_64;
1.531 + step2[0] = dct_const_round_shift(temp1);
1.532 + step2[1] = dct_const_round_shift(temp2);
1.533 + temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64;
1.534 + temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64;
1.535 + step2[2] = dct_const_round_shift(temp1);
1.536 + step2[3] = dct_const_round_shift(temp2);
1.537 + step2[4] = step1[4] + step1[5];
1.538 + step2[5] = step1[4] - step1[5];
1.539 + step2[6] = -step1[6] + step1[7];
1.540 + step2[7] = step1[6] + step1[7];
1.541 +
1.542 + step2[8] = step1[8];
1.543 + step2[15] = step1[15];
1.544 + temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64;
1.545 + temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64;
1.546 + step2[9] = dct_const_round_shift(temp1);
1.547 + step2[14] = dct_const_round_shift(temp2);
1.548 + temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64;
1.549 + temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64;
1.550 + step2[10] = dct_const_round_shift(temp1);
1.551 + step2[13] = dct_const_round_shift(temp2);
1.552 + step2[11] = step1[11];
1.553 + step2[12] = step1[12];
1.554 +
1.555 + // stage 5
1.556 + step1[0] = step2[0] + step2[3];
1.557 + step1[1] = step2[1] + step2[2];
1.558 + step1[2] = step2[1] - step2[2];
1.559 + step1[3] = step2[0] - step2[3];
1.560 + step1[4] = step2[4];
1.561 + temp1 = (step2[6] - step2[5]) * cospi_16_64;
1.562 + temp2 = (step2[5] + step2[6]) * cospi_16_64;
1.563 + step1[5] = dct_const_round_shift(temp1);
1.564 + step1[6] = dct_const_round_shift(temp2);
1.565 + step1[7] = step2[7];
1.566 +
1.567 + step1[8] = step2[8] + step2[11];
1.568 + step1[9] = step2[9] + step2[10];
1.569 + step1[10] = step2[9] - step2[10];
1.570 + step1[11] = step2[8] - step2[11];
1.571 + step1[12] = -step2[12] + step2[15];
1.572 + step1[13] = -step2[13] + step2[14];
1.573 + step1[14] = step2[13] + step2[14];
1.574 + step1[15] = step2[12] + step2[15];
1.575 +
1.576 + // stage 6
1.577 + step2[0] = step1[0] + step1[7];
1.578 + step2[1] = step1[1] + step1[6];
1.579 + step2[2] = step1[2] + step1[5];
1.580 + step2[3] = step1[3] + step1[4];
1.581 + step2[4] = step1[3] - step1[4];
1.582 + step2[5] = step1[2] - step1[5];
1.583 + step2[6] = step1[1] - step1[6];
1.584 + step2[7] = step1[0] - step1[7];
1.585 + step2[8] = step1[8];
1.586 + step2[9] = step1[9];
1.587 + temp1 = (-step1[10] + step1[13]) * cospi_16_64;
1.588 + temp2 = (step1[10] + step1[13]) * cospi_16_64;
1.589 + step2[10] = dct_const_round_shift(temp1);
1.590 + step2[13] = dct_const_round_shift(temp2);
1.591 + temp1 = (-step1[11] + step1[12]) * cospi_16_64;
1.592 + temp2 = (step1[11] + step1[12]) * cospi_16_64;
1.593 + step2[11] = dct_const_round_shift(temp1);
1.594 + step2[12] = dct_const_round_shift(temp2);
1.595 + step2[14] = step1[14];
1.596 + step2[15] = step1[15];
1.597 +
1.598 + // stage 7
1.599 + output[0] = step2[0] + step2[15];
1.600 + output[1] = step2[1] + step2[14];
1.601 + output[2] = step2[2] + step2[13];
1.602 + output[3] = step2[3] + step2[12];
1.603 + output[4] = step2[4] + step2[11];
1.604 + output[5] = step2[5] + step2[10];
1.605 + output[6] = step2[6] + step2[9];
1.606 + output[7] = step2[7] + step2[8];
1.607 + output[8] = step2[7] - step2[8];
1.608 + output[9] = step2[6] - step2[9];
1.609 + output[10] = step2[5] - step2[10];
1.610 + output[11] = step2[4] - step2[11];
1.611 + output[12] = step2[3] - step2[12];
1.612 + output[13] = step2[2] - step2[13];
1.613 + output[14] = step2[1] - step2[14];
1.614 + output[15] = step2[0] - step2[15];
1.615 +}
1.616 +
1.617 +void vp9_idct16x16_256_add_c(const int16_t *input, uint8_t *dest, int stride) {
1.618 + int16_t out[16 * 16];
1.619 + int16_t *outptr = out;
1.620 + int i, j;
1.621 + int16_t temp_in[16], temp_out[16];
1.622 +
1.623 + // First transform rows
1.624 + for (i = 0; i < 16; ++i) {
1.625 + idct16_1d(input, outptr);
1.626 + input += 16;
1.627 + outptr += 16;
1.628 + }
1.629 +
1.630 + // Then transform columns
1.631 + for (i = 0; i < 16; ++i) {
1.632 + for (j = 0; j < 16; ++j)
1.633 + temp_in[j] = out[j * 16 + i];
1.634 + idct16_1d(temp_in, temp_out);
1.635 + for (j = 0; j < 16; ++j)
1.636 + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
1.637 + + dest[j * stride + i]);
1.638 + }
1.639 +}
1.640 +
1.641 +static void iadst16_1d(const int16_t *input, int16_t *output) {
1.642 + int s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15;
1.643 +
1.644 + int x0 = input[15];
1.645 + int x1 = input[0];
1.646 + int x2 = input[13];
1.647 + int x3 = input[2];
1.648 + int x4 = input[11];
1.649 + int x5 = input[4];
1.650 + int x6 = input[9];
1.651 + int x7 = input[6];
1.652 + int x8 = input[7];
1.653 + int x9 = input[8];
1.654 + int x10 = input[5];
1.655 + int x11 = input[10];
1.656 + int x12 = input[3];
1.657 + int x13 = input[12];
1.658 + int x14 = input[1];
1.659 + int x15 = input[14];
1.660 +
1.661 + if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7 | x8
1.662 + | x9 | x10 | x11 | x12 | x13 | x14 | x15)) {
1.663 + output[0] = output[1] = output[2] = output[3] = output[4]
1.664 + = output[5] = output[6] = output[7] = output[8]
1.665 + = output[9] = output[10] = output[11] = output[12]
1.666 + = output[13] = output[14] = output[15] = 0;
1.667 + return;
1.668 + }
1.669 +
1.670 + // stage 1
1.671 + s0 = x0 * cospi_1_64 + x1 * cospi_31_64;
1.672 + s1 = x0 * cospi_31_64 - x1 * cospi_1_64;
1.673 + s2 = x2 * cospi_5_64 + x3 * cospi_27_64;
1.674 + s3 = x2 * cospi_27_64 - x3 * cospi_5_64;
1.675 + s4 = x4 * cospi_9_64 + x5 * cospi_23_64;
1.676 + s5 = x4 * cospi_23_64 - x5 * cospi_9_64;
1.677 + s6 = x6 * cospi_13_64 + x7 * cospi_19_64;
1.678 + s7 = x6 * cospi_19_64 - x7 * cospi_13_64;
1.679 + s8 = x8 * cospi_17_64 + x9 * cospi_15_64;
1.680 + s9 = x8 * cospi_15_64 - x9 * cospi_17_64;
1.681 + s10 = x10 * cospi_21_64 + x11 * cospi_11_64;
1.682 + s11 = x10 * cospi_11_64 - x11 * cospi_21_64;
1.683 + s12 = x12 * cospi_25_64 + x13 * cospi_7_64;
1.684 + s13 = x12 * cospi_7_64 - x13 * cospi_25_64;
1.685 + s14 = x14 * cospi_29_64 + x15 * cospi_3_64;
1.686 + s15 = x14 * cospi_3_64 - x15 * cospi_29_64;
1.687 +
1.688 + x0 = dct_const_round_shift(s0 + s8);
1.689 + x1 = dct_const_round_shift(s1 + s9);
1.690 + x2 = dct_const_round_shift(s2 + s10);
1.691 + x3 = dct_const_round_shift(s3 + s11);
1.692 + x4 = dct_const_round_shift(s4 + s12);
1.693 + x5 = dct_const_round_shift(s5 + s13);
1.694 + x6 = dct_const_round_shift(s6 + s14);
1.695 + x7 = dct_const_round_shift(s7 + s15);
1.696 + x8 = dct_const_round_shift(s0 - s8);
1.697 + x9 = dct_const_round_shift(s1 - s9);
1.698 + x10 = dct_const_round_shift(s2 - s10);
1.699 + x11 = dct_const_round_shift(s3 - s11);
1.700 + x12 = dct_const_round_shift(s4 - s12);
1.701 + x13 = dct_const_round_shift(s5 - s13);
1.702 + x14 = dct_const_round_shift(s6 - s14);
1.703 + x15 = dct_const_round_shift(s7 - s15);
1.704 +
1.705 + // stage 2
1.706 + s0 = x0;
1.707 + s1 = x1;
1.708 + s2 = x2;
1.709 + s3 = x3;
1.710 + s4 = x4;
1.711 + s5 = x5;
1.712 + s6 = x6;
1.713 + s7 = x7;
1.714 + s8 = x8 * cospi_4_64 + x9 * cospi_28_64;
1.715 + s9 = x8 * cospi_28_64 - x9 * cospi_4_64;
1.716 + s10 = x10 * cospi_20_64 + x11 * cospi_12_64;
1.717 + s11 = x10 * cospi_12_64 - x11 * cospi_20_64;
1.718 + s12 = - x12 * cospi_28_64 + x13 * cospi_4_64;
1.719 + s13 = x12 * cospi_4_64 + x13 * cospi_28_64;
1.720 + s14 = - x14 * cospi_12_64 + x15 * cospi_20_64;
1.721 + s15 = x14 * cospi_20_64 + x15 * cospi_12_64;
1.722 +
1.723 + x0 = s0 + s4;
1.724 + x1 = s1 + s5;
1.725 + x2 = s2 + s6;
1.726 + x3 = s3 + s7;
1.727 + x4 = s0 - s4;
1.728 + x5 = s1 - s5;
1.729 + x6 = s2 - s6;
1.730 + x7 = s3 - s7;
1.731 + x8 = dct_const_round_shift(s8 + s12);
1.732 + x9 = dct_const_round_shift(s9 + s13);
1.733 + x10 = dct_const_round_shift(s10 + s14);
1.734 + x11 = dct_const_round_shift(s11 + s15);
1.735 + x12 = dct_const_round_shift(s8 - s12);
1.736 + x13 = dct_const_round_shift(s9 - s13);
1.737 + x14 = dct_const_round_shift(s10 - s14);
1.738 + x15 = dct_const_round_shift(s11 - s15);
1.739 +
1.740 + // stage 3
1.741 + s0 = x0;
1.742 + s1 = x1;
1.743 + s2 = x2;
1.744 + s3 = x3;
1.745 + s4 = x4 * cospi_8_64 + x5 * cospi_24_64;
1.746 + s5 = x4 * cospi_24_64 - x5 * cospi_8_64;
1.747 + s6 = - x6 * cospi_24_64 + x7 * cospi_8_64;
1.748 + s7 = x6 * cospi_8_64 + x7 * cospi_24_64;
1.749 + s8 = x8;
1.750 + s9 = x9;
1.751 + s10 = x10;
1.752 + s11 = x11;
1.753 + s12 = x12 * cospi_8_64 + x13 * cospi_24_64;
1.754 + s13 = x12 * cospi_24_64 - x13 * cospi_8_64;
1.755 + s14 = - x14 * cospi_24_64 + x15 * cospi_8_64;
1.756 + s15 = x14 * cospi_8_64 + x15 * cospi_24_64;
1.757 +
1.758 + x0 = s0 + s2;
1.759 + x1 = s1 + s3;
1.760 + x2 = s0 - s2;
1.761 + x3 = s1 - s3;
1.762 + x4 = dct_const_round_shift(s4 + s6);
1.763 + x5 = dct_const_round_shift(s5 + s7);
1.764 + x6 = dct_const_round_shift(s4 - s6);
1.765 + x7 = dct_const_round_shift(s5 - s7);
1.766 + x8 = s8 + s10;
1.767 + x9 = s9 + s11;
1.768 + x10 = s8 - s10;
1.769 + x11 = s9 - s11;
1.770 + x12 = dct_const_round_shift(s12 + s14);
1.771 + x13 = dct_const_round_shift(s13 + s15);
1.772 + x14 = dct_const_round_shift(s12 - s14);
1.773 + x15 = dct_const_round_shift(s13 - s15);
1.774 +
1.775 + // stage 4
1.776 + s2 = (- cospi_16_64) * (x2 + x3);
1.777 + s3 = cospi_16_64 * (x2 - x3);
1.778 + s6 = cospi_16_64 * (x6 + x7);
1.779 + s7 = cospi_16_64 * (- x6 + x7);
1.780 + s10 = cospi_16_64 * (x10 + x11);
1.781 + s11 = cospi_16_64 * (- x10 + x11);
1.782 + s14 = (- cospi_16_64) * (x14 + x15);
1.783 + s15 = cospi_16_64 * (x14 - x15);
1.784 +
1.785 + x2 = dct_const_round_shift(s2);
1.786 + x3 = dct_const_round_shift(s3);
1.787 + x6 = dct_const_round_shift(s6);
1.788 + x7 = dct_const_round_shift(s7);
1.789 + x10 = dct_const_round_shift(s10);
1.790 + x11 = dct_const_round_shift(s11);
1.791 + x14 = dct_const_round_shift(s14);
1.792 + x15 = dct_const_round_shift(s15);
1.793 +
1.794 + output[0] = x0;
1.795 + output[1] = -x8;
1.796 + output[2] = x12;
1.797 + output[3] = -x4;
1.798 + output[4] = x6;
1.799 + output[5] = x14;
1.800 + output[6] = x10;
1.801 + output[7] = x2;
1.802 + output[8] = x3;
1.803 + output[9] = x11;
1.804 + output[10] = x15;
1.805 + output[11] = x7;
1.806 + output[12] = x5;
1.807 + output[13] = -x13;
1.808 + output[14] = x9;
1.809 + output[15] = -x1;
1.810 +}
1.811 +
1.812 +static const transform_2d IHT_16[] = {
1.813 + { idct16_1d, idct16_1d }, // DCT_DCT = 0
1.814 + { iadst16_1d, idct16_1d }, // ADST_DCT = 1
1.815 + { idct16_1d, iadst16_1d }, // DCT_ADST = 2
1.816 + { iadst16_1d, iadst16_1d } // ADST_ADST = 3
1.817 +};
1.818 +
1.819 +void vp9_iht16x16_256_add_c(const int16_t *input, uint8_t *dest, int stride,
1.820 + int tx_type) {
1.821 + int i, j;
1.822 + int16_t out[16 * 16];
1.823 + int16_t *outptr = out;
1.824 + int16_t temp_in[16], temp_out[16];
1.825 + const transform_2d ht = IHT_16[tx_type];
1.826 +
1.827 + // Rows
1.828 + for (i = 0; i < 16; ++i) {
1.829 + ht.rows(input, outptr);
1.830 + input += 16;
1.831 + outptr += 16;
1.832 + }
1.833 +
1.834 + // Columns
1.835 + for (i = 0; i < 16; ++i) {
1.836 + for (j = 0; j < 16; ++j)
1.837 + temp_in[j] = out[j * 16 + i];
1.838 + ht.cols(temp_in, temp_out);
1.839 + for (j = 0; j < 16; ++j)
1.840 + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
1.841 + + dest[j * stride + i]);
1.842 + }
1.843 +}
1.844 +
1.845 +void vp9_idct16x16_10_add_c(const int16_t *input, uint8_t *dest, int stride) {
1.846 + int16_t out[16 * 16] = { 0 };
1.847 + int16_t *outptr = out;
1.848 + int i, j;
1.849 + int16_t temp_in[16], temp_out[16];
1.850 +
1.851 + // First transform rows. Since all non-zero dct coefficients are in
1.852 + // upper-left 4x4 area, we only need to calculate first 4 rows here.
1.853 + for (i = 0; i < 4; ++i) {
1.854 + idct16_1d(input, outptr);
1.855 + input += 16;
1.856 + outptr += 16;
1.857 + }
1.858 +
1.859 + // Then transform columns
1.860 + for (i = 0; i < 16; ++i) {
1.861 + for (j = 0; j < 16; ++j)
1.862 + temp_in[j] = out[j*16 + i];
1.863 + idct16_1d(temp_in, temp_out);
1.864 + for (j = 0; j < 16; ++j)
1.865 + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
1.866 + + dest[j * stride + i]);
1.867 + }
1.868 +}
1.869 +
1.870 +void vp9_idct16x16_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
1.871 + int i, j;
1.872 + int a1;
1.873 + int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
1.874 + out = dct_const_round_shift(out * cospi_16_64);
1.875 + a1 = ROUND_POWER_OF_TWO(out, 6);
1.876 + for (j = 0; j < 16; ++j) {
1.877 + for (i = 0; i < 16; ++i)
1.878 + dest[i] = clip_pixel(dest[i] + a1);
1.879 + dest += stride;
1.880 + }
1.881 +}
1.882 +
1.883 +static void idct32_1d(const int16_t *input, int16_t *output) {
1.884 + int16_t step1[32], step2[32];
1.885 + int temp1, temp2;
1.886 +
1.887 + // stage 1
1.888 + step1[0] = input[0];
1.889 + step1[1] = input[16];
1.890 + step1[2] = input[8];
1.891 + step1[3] = input[24];
1.892 + step1[4] = input[4];
1.893 + step1[5] = input[20];
1.894 + step1[6] = input[12];
1.895 + step1[7] = input[28];
1.896 + step1[8] = input[2];
1.897 + step1[9] = input[18];
1.898 + step1[10] = input[10];
1.899 + step1[11] = input[26];
1.900 + step1[12] = input[6];
1.901 + step1[13] = input[22];
1.902 + step1[14] = input[14];
1.903 + step1[15] = input[30];
1.904 +
1.905 + temp1 = input[1] * cospi_31_64 - input[31] * cospi_1_64;
1.906 + temp2 = input[1] * cospi_1_64 + input[31] * cospi_31_64;
1.907 + step1[16] = dct_const_round_shift(temp1);
1.908 + step1[31] = dct_const_round_shift(temp2);
1.909 +
1.910 + temp1 = input[17] * cospi_15_64 - input[15] * cospi_17_64;
1.911 + temp2 = input[17] * cospi_17_64 + input[15] * cospi_15_64;
1.912 + step1[17] = dct_const_round_shift(temp1);
1.913 + step1[30] = dct_const_round_shift(temp2);
1.914 +
1.915 + temp1 = input[9] * cospi_23_64 - input[23] * cospi_9_64;
1.916 + temp2 = input[9] * cospi_9_64 + input[23] * cospi_23_64;
1.917 + step1[18] = dct_const_round_shift(temp1);
1.918 + step1[29] = dct_const_round_shift(temp2);
1.919 +
1.920 + temp1 = input[25] * cospi_7_64 - input[7] * cospi_25_64;
1.921 + temp2 = input[25] * cospi_25_64 + input[7] * cospi_7_64;
1.922 + step1[19] = dct_const_round_shift(temp1);
1.923 + step1[28] = dct_const_round_shift(temp2);
1.924 +
1.925 + temp1 = input[5] * cospi_27_64 - input[27] * cospi_5_64;
1.926 + temp2 = input[5] * cospi_5_64 + input[27] * cospi_27_64;
1.927 + step1[20] = dct_const_round_shift(temp1);
1.928 + step1[27] = dct_const_round_shift(temp2);
1.929 +
1.930 + temp1 = input[21] * cospi_11_64 - input[11] * cospi_21_64;
1.931 + temp2 = input[21] * cospi_21_64 + input[11] * cospi_11_64;
1.932 + step1[21] = dct_const_round_shift(temp1);
1.933 + step1[26] = dct_const_round_shift(temp2);
1.934 +
1.935 + temp1 = input[13] * cospi_19_64 - input[19] * cospi_13_64;
1.936 + temp2 = input[13] * cospi_13_64 + input[19] * cospi_19_64;
1.937 + step1[22] = dct_const_round_shift(temp1);
1.938 + step1[25] = dct_const_round_shift(temp2);
1.939 +
1.940 + temp1 = input[29] * cospi_3_64 - input[3] * cospi_29_64;
1.941 + temp2 = input[29] * cospi_29_64 + input[3] * cospi_3_64;
1.942 + step1[23] = dct_const_round_shift(temp1);
1.943 + step1[24] = dct_const_round_shift(temp2);
1.944 +
1.945 + // stage 2
1.946 + step2[0] = step1[0];
1.947 + step2[1] = step1[1];
1.948 + step2[2] = step1[2];
1.949 + step2[3] = step1[3];
1.950 + step2[4] = step1[4];
1.951 + step2[5] = step1[5];
1.952 + step2[6] = step1[6];
1.953 + step2[7] = step1[7];
1.954 +
1.955 + temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64;
1.956 + temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64;
1.957 + step2[8] = dct_const_round_shift(temp1);
1.958 + step2[15] = dct_const_round_shift(temp2);
1.959 +
1.960 + temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64;
1.961 + temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64;
1.962 + step2[9] = dct_const_round_shift(temp1);
1.963 + step2[14] = dct_const_round_shift(temp2);
1.964 +
1.965 + temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64;
1.966 + temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64;
1.967 + step2[10] = dct_const_round_shift(temp1);
1.968 + step2[13] = dct_const_round_shift(temp2);
1.969 +
1.970 + temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64;
1.971 + temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64;
1.972 + step2[11] = dct_const_round_shift(temp1);
1.973 + step2[12] = dct_const_round_shift(temp2);
1.974 +
1.975 + step2[16] = step1[16] + step1[17];
1.976 + step2[17] = step1[16] - step1[17];
1.977 + step2[18] = -step1[18] + step1[19];
1.978 + step2[19] = step1[18] + step1[19];
1.979 + step2[20] = step1[20] + step1[21];
1.980 + step2[21] = step1[20] - step1[21];
1.981 + step2[22] = -step1[22] + step1[23];
1.982 + step2[23] = step1[22] + step1[23];
1.983 + step2[24] = step1[24] + step1[25];
1.984 + step2[25] = step1[24] - step1[25];
1.985 + step2[26] = -step1[26] + step1[27];
1.986 + step2[27] = step1[26] + step1[27];
1.987 + step2[28] = step1[28] + step1[29];
1.988 + step2[29] = step1[28] - step1[29];
1.989 + step2[30] = -step1[30] + step1[31];
1.990 + step2[31] = step1[30] + step1[31];
1.991 +
1.992 + // stage 3
1.993 + step1[0] = step2[0];
1.994 + step1[1] = step2[1];
1.995 + step1[2] = step2[2];
1.996 + step1[3] = step2[3];
1.997 +
1.998 + temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64;
1.999 + temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64;
1.1000 + step1[4] = dct_const_round_shift(temp1);
1.1001 + step1[7] = dct_const_round_shift(temp2);
1.1002 + temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64;
1.1003 + temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64;
1.1004 + step1[5] = dct_const_round_shift(temp1);
1.1005 + step1[6] = dct_const_round_shift(temp2);
1.1006 +
1.1007 + step1[8] = step2[8] + step2[9];
1.1008 + step1[9] = step2[8] - step2[9];
1.1009 + step1[10] = -step2[10] + step2[11];
1.1010 + step1[11] = step2[10] + step2[11];
1.1011 + step1[12] = step2[12] + step2[13];
1.1012 + step1[13] = step2[12] - step2[13];
1.1013 + step1[14] = -step2[14] + step2[15];
1.1014 + step1[15] = step2[14] + step2[15];
1.1015 +
1.1016 + step1[16] = step2[16];
1.1017 + step1[31] = step2[31];
1.1018 + temp1 = -step2[17] * cospi_4_64 + step2[30] * cospi_28_64;
1.1019 + temp2 = step2[17] * cospi_28_64 + step2[30] * cospi_4_64;
1.1020 + step1[17] = dct_const_round_shift(temp1);
1.1021 + step1[30] = dct_const_round_shift(temp2);
1.1022 + temp1 = -step2[18] * cospi_28_64 - step2[29] * cospi_4_64;
1.1023 + temp2 = -step2[18] * cospi_4_64 + step2[29] * cospi_28_64;
1.1024 + step1[18] = dct_const_round_shift(temp1);
1.1025 + step1[29] = dct_const_round_shift(temp2);
1.1026 + step1[19] = step2[19];
1.1027 + step1[20] = step2[20];
1.1028 + temp1 = -step2[21] * cospi_20_64 + step2[26] * cospi_12_64;
1.1029 + temp2 = step2[21] * cospi_12_64 + step2[26] * cospi_20_64;
1.1030 + step1[21] = dct_const_round_shift(temp1);
1.1031 + step1[26] = dct_const_round_shift(temp2);
1.1032 + temp1 = -step2[22] * cospi_12_64 - step2[25] * cospi_20_64;
1.1033 + temp2 = -step2[22] * cospi_20_64 + step2[25] * cospi_12_64;
1.1034 + step1[22] = dct_const_round_shift(temp1);
1.1035 + step1[25] = dct_const_round_shift(temp2);
1.1036 + step1[23] = step2[23];
1.1037 + step1[24] = step2[24];
1.1038 + step1[27] = step2[27];
1.1039 + step1[28] = step2[28];
1.1040 +
1.1041 + // stage 4
1.1042 + temp1 = (step1[0] + step1[1]) * cospi_16_64;
1.1043 + temp2 = (step1[0] - step1[1]) * cospi_16_64;
1.1044 + step2[0] = dct_const_round_shift(temp1);
1.1045 + step2[1] = dct_const_round_shift(temp2);
1.1046 + temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64;
1.1047 + temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64;
1.1048 + step2[2] = dct_const_round_shift(temp1);
1.1049 + step2[3] = dct_const_round_shift(temp2);
1.1050 + step2[4] = step1[4] + step1[5];
1.1051 + step2[5] = step1[4] - step1[5];
1.1052 + step2[6] = -step1[6] + step1[7];
1.1053 + step2[7] = step1[6] + step1[7];
1.1054 +
1.1055 + step2[8] = step1[8];
1.1056 + step2[15] = step1[15];
1.1057 + temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64;
1.1058 + temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64;
1.1059 + step2[9] = dct_const_round_shift(temp1);
1.1060 + step2[14] = dct_const_round_shift(temp2);
1.1061 + temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64;
1.1062 + temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64;
1.1063 + step2[10] = dct_const_round_shift(temp1);
1.1064 + step2[13] = dct_const_round_shift(temp2);
1.1065 + step2[11] = step1[11];
1.1066 + step2[12] = step1[12];
1.1067 +
1.1068 + step2[16] = step1[16] + step1[19];
1.1069 + step2[17] = step1[17] + step1[18];
1.1070 + step2[18] = step1[17] - step1[18];
1.1071 + step2[19] = step1[16] - step1[19];
1.1072 + step2[20] = -step1[20] + step1[23];
1.1073 + step2[21] = -step1[21] + step1[22];
1.1074 + step2[22] = step1[21] + step1[22];
1.1075 + step2[23] = step1[20] + step1[23];
1.1076 +
1.1077 + step2[24] = step1[24] + step1[27];
1.1078 + step2[25] = step1[25] + step1[26];
1.1079 + step2[26] = step1[25] - step1[26];
1.1080 + step2[27] = step1[24] - step1[27];
1.1081 + step2[28] = -step1[28] + step1[31];
1.1082 + step2[29] = -step1[29] + step1[30];
1.1083 + step2[30] = step1[29] + step1[30];
1.1084 + step2[31] = step1[28] + step1[31];
1.1085 +
1.1086 + // stage 5
1.1087 + step1[0] = step2[0] + step2[3];
1.1088 + step1[1] = step2[1] + step2[2];
1.1089 + step1[2] = step2[1] - step2[2];
1.1090 + step1[3] = step2[0] - step2[3];
1.1091 + step1[4] = step2[4];
1.1092 + temp1 = (step2[6] - step2[5]) * cospi_16_64;
1.1093 + temp2 = (step2[5] + step2[6]) * cospi_16_64;
1.1094 + step1[5] = dct_const_round_shift(temp1);
1.1095 + step1[6] = dct_const_round_shift(temp2);
1.1096 + step1[7] = step2[7];
1.1097 +
1.1098 + step1[8] = step2[8] + step2[11];
1.1099 + step1[9] = step2[9] + step2[10];
1.1100 + step1[10] = step2[9] - step2[10];
1.1101 + step1[11] = step2[8] - step2[11];
1.1102 + step1[12] = -step2[12] + step2[15];
1.1103 + step1[13] = -step2[13] + step2[14];
1.1104 + step1[14] = step2[13] + step2[14];
1.1105 + step1[15] = step2[12] + step2[15];
1.1106 +
1.1107 + step1[16] = step2[16];
1.1108 + step1[17] = step2[17];
1.1109 + temp1 = -step2[18] * cospi_8_64 + step2[29] * cospi_24_64;
1.1110 + temp2 = step2[18] * cospi_24_64 + step2[29] * cospi_8_64;
1.1111 + step1[18] = dct_const_round_shift(temp1);
1.1112 + step1[29] = dct_const_round_shift(temp2);
1.1113 + temp1 = -step2[19] * cospi_8_64 + step2[28] * cospi_24_64;
1.1114 + temp2 = step2[19] * cospi_24_64 + step2[28] * cospi_8_64;
1.1115 + step1[19] = dct_const_round_shift(temp1);
1.1116 + step1[28] = dct_const_round_shift(temp2);
1.1117 + temp1 = -step2[20] * cospi_24_64 - step2[27] * cospi_8_64;
1.1118 + temp2 = -step2[20] * cospi_8_64 + step2[27] * cospi_24_64;
1.1119 + step1[20] = dct_const_round_shift(temp1);
1.1120 + step1[27] = dct_const_round_shift(temp2);
1.1121 + temp1 = -step2[21] * cospi_24_64 - step2[26] * cospi_8_64;
1.1122 + temp2 = -step2[21] * cospi_8_64 + step2[26] * cospi_24_64;
1.1123 + step1[21] = dct_const_round_shift(temp1);
1.1124 + step1[26] = dct_const_round_shift(temp2);
1.1125 + step1[22] = step2[22];
1.1126 + step1[23] = step2[23];
1.1127 + step1[24] = step2[24];
1.1128 + step1[25] = step2[25];
1.1129 + step1[30] = step2[30];
1.1130 + step1[31] = step2[31];
1.1131 +
1.1132 + // stage 6
1.1133 + step2[0] = step1[0] + step1[7];
1.1134 + step2[1] = step1[1] + step1[6];
1.1135 + step2[2] = step1[2] + step1[5];
1.1136 + step2[3] = step1[3] + step1[4];
1.1137 + step2[4] = step1[3] - step1[4];
1.1138 + step2[5] = step1[2] - step1[5];
1.1139 + step2[6] = step1[1] - step1[6];
1.1140 + step2[7] = step1[0] - step1[7];
1.1141 + step2[8] = step1[8];
1.1142 + step2[9] = step1[9];
1.1143 + temp1 = (-step1[10] + step1[13]) * cospi_16_64;
1.1144 + temp2 = (step1[10] + step1[13]) * cospi_16_64;
1.1145 + step2[10] = dct_const_round_shift(temp1);
1.1146 + step2[13] = dct_const_round_shift(temp2);
1.1147 + temp1 = (-step1[11] + step1[12]) * cospi_16_64;
1.1148 + temp2 = (step1[11] + step1[12]) * cospi_16_64;
1.1149 + step2[11] = dct_const_round_shift(temp1);
1.1150 + step2[12] = dct_const_round_shift(temp2);
1.1151 + step2[14] = step1[14];
1.1152 + step2[15] = step1[15];
1.1153 +
1.1154 + step2[16] = step1[16] + step1[23];
1.1155 + step2[17] = step1[17] + step1[22];
1.1156 + step2[18] = step1[18] + step1[21];
1.1157 + step2[19] = step1[19] + step1[20];
1.1158 + step2[20] = step1[19] - step1[20];
1.1159 + step2[21] = step1[18] - step1[21];
1.1160 + step2[22] = step1[17] - step1[22];
1.1161 + step2[23] = step1[16] - step1[23];
1.1162 +
1.1163 + step2[24] = -step1[24] + step1[31];
1.1164 + step2[25] = -step1[25] + step1[30];
1.1165 + step2[26] = -step1[26] + step1[29];
1.1166 + step2[27] = -step1[27] + step1[28];
1.1167 + step2[28] = step1[27] + step1[28];
1.1168 + step2[29] = step1[26] + step1[29];
1.1169 + step2[30] = step1[25] + step1[30];
1.1170 + step2[31] = step1[24] + step1[31];
1.1171 +
1.1172 + // stage 7
1.1173 + step1[0] = step2[0] + step2[15];
1.1174 + step1[1] = step2[1] + step2[14];
1.1175 + step1[2] = step2[2] + step2[13];
1.1176 + step1[3] = step2[3] + step2[12];
1.1177 + step1[4] = step2[4] + step2[11];
1.1178 + step1[5] = step2[5] + step2[10];
1.1179 + step1[6] = step2[6] + step2[9];
1.1180 + step1[7] = step2[7] + step2[8];
1.1181 + step1[8] = step2[7] - step2[8];
1.1182 + step1[9] = step2[6] - step2[9];
1.1183 + step1[10] = step2[5] - step2[10];
1.1184 + step1[11] = step2[4] - step2[11];
1.1185 + step1[12] = step2[3] - step2[12];
1.1186 + step1[13] = step2[2] - step2[13];
1.1187 + step1[14] = step2[1] - step2[14];
1.1188 + step1[15] = step2[0] - step2[15];
1.1189 +
1.1190 + step1[16] = step2[16];
1.1191 + step1[17] = step2[17];
1.1192 + step1[18] = step2[18];
1.1193 + step1[19] = step2[19];
1.1194 + temp1 = (-step2[20] + step2[27]) * cospi_16_64;
1.1195 + temp2 = (step2[20] + step2[27]) * cospi_16_64;
1.1196 + step1[20] = dct_const_round_shift(temp1);
1.1197 + step1[27] = dct_const_round_shift(temp2);
1.1198 + temp1 = (-step2[21] + step2[26]) * cospi_16_64;
1.1199 + temp2 = (step2[21] + step2[26]) * cospi_16_64;
1.1200 + step1[21] = dct_const_round_shift(temp1);
1.1201 + step1[26] = dct_const_round_shift(temp2);
1.1202 + temp1 = (-step2[22] + step2[25]) * cospi_16_64;
1.1203 + temp2 = (step2[22] + step2[25]) * cospi_16_64;
1.1204 + step1[22] = dct_const_round_shift(temp1);
1.1205 + step1[25] = dct_const_round_shift(temp2);
1.1206 + temp1 = (-step2[23] + step2[24]) * cospi_16_64;
1.1207 + temp2 = (step2[23] + step2[24]) * cospi_16_64;
1.1208 + step1[23] = dct_const_round_shift(temp1);
1.1209 + step1[24] = dct_const_round_shift(temp2);
1.1210 + step1[28] = step2[28];
1.1211 + step1[29] = step2[29];
1.1212 + step1[30] = step2[30];
1.1213 + step1[31] = step2[31];
1.1214 +
1.1215 + // final stage
1.1216 + output[0] = step1[0] + step1[31];
1.1217 + output[1] = step1[1] + step1[30];
1.1218 + output[2] = step1[2] + step1[29];
1.1219 + output[3] = step1[3] + step1[28];
1.1220 + output[4] = step1[4] + step1[27];
1.1221 + output[5] = step1[5] + step1[26];
1.1222 + output[6] = step1[6] + step1[25];
1.1223 + output[7] = step1[7] + step1[24];
1.1224 + output[8] = step1[8] + step1[23];
1.1225 + output[9] = step1[9] + step1[22];
1.1226 + output[10] = step1[10] + step1[21];
1.1227 + output[11] = step1[11] + step1[20];
1.1228 + output[12] = step1[12] + step1[19];
1.1229 + output[13] = step1[13] + step1[18];
1.1230 + output[14] = step1[14] + step1[17];
1.1231 + output[15] = step1[15] + step1[16];
1.1232 + output[16] = step1[15] - step1[16];
1.1233 + output[17] = step1[14] - step1[17];
1.1234 + output[18] = step1[13] - step1[18];
1.1235 + output[19] = step1[12] - step1[19];
1.1236 + output[20] = step1[11] - step1[20];
1.1237 + output[21] = step1[10] - step1[21];
1.1238 + output[22] = step1[9] - step1[22];
1.1239 + output[23] = step1[8] - step1[23];
1.1240 + output[24] = step1[7] - step1[24];
1.1241 + output[25] = step1[6] - step1[25];
1.1242 + output[26] = step1[5] - step1[26];
1.1243 + output[27] = step1[4] - step1[27];
1.1244 + output[28] = step1[3] - step1[28];
1.1245 + output[29] = step1[2] - step1[29];
1.1246 + output[30] = step1[1] - step1[30];
1.1247 + output[31] = step1[0] - step1[31];
1.1248 +}
1.1249 +
1.1250 +void vp9_idct32x32_1024_add_c(const int16_t *input, uint8_t *dest, int stride) {
1.1251 + int16_t out[32 * 32];
1.1252 + int16_t *outptr = out;
1.1253 + int i, j;
1.1254 + int16_t temp_in[32], temp_out[32];
1.1255 +
1.1256 + // Rows
1.1257 + for (i = 0; i < 32; ++i) {
1.1258 + int16_t zero_coeff[16];
1.1259 + for (j = 0; j < 16; ++j)
1.1260 + zero_coeff[j] = input[2 * j] | input[2 * j + 1];
1.1261 + for (j = 0; j < 8; ++j)
1.1262 + zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
1.1263 + for (j = 0; j < 4; ++j)
1.1264 + zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
1.1265 + for (j = 0; j < 2; ++j)
1.1266 + zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
1.1267 +
1.1268 + if (zero_coeff[0] | zero_coeff[1])
1.1269 + idct32_1d(input, outptr);
1.1270 + else
1.1271 + vpx_memset(outptr, 0, sizeof(int16_t) * 32);
1.1272 + input += 32;
1.1273 + outptr += 32;
1.1274 + }
1.1275 +
1.1276 + // Columns
1.1277 + for (i = 0; i < 32; ++i) {
1.1278 + for (j = 0; j < 32; ++j)
1.1279 + temp_in[j] = out[j * 32 + i];
1.1280 + idct32_1d(temp_in, temp_out);
1.1281 + for (j = 0; j < 32; ++j)
1.1282 + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
1.1283 + + dest[j * stride + i]);
1.1284 + }
1.1285 +}
1.1286 +
1.1287 +void vp9_idct32x32_34_add_c(const int16_t *input, uint8_t *dest, int stride) {
1.1288 + int16_t out[32 * 32] = {0};
1.1289 + int16_t *outptr = out;
1.1290 + int i, j;
1.1291 + int16_t temp_in[32], temp_out[32];
1.1292 +
1.1293 + // Rows
1.1294 + // only upper-left 8x8 has non-zero coeff
1.1295 + for (i = 0; i < 8; ++i) {
1.1296 + idct32_1d(input, outptr);
1.1297 + input += 32;
1.1298 + outptr += 32;
1.1299 + }
1.1300 +
1.1301 + // Columns
1.1302 + for (i = 0; i < 32; ++i) {
1.1303 + for (j = 0; j < 32; ++j)
1.1304 + temp_in[j] = out[j * 32 + i];
1.1305 + idct32_1d(temp_in, temp_out);
1.1306 + for (j = 0; j < 32; ++j)
1.1307 + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
1.1308 + + dest[j * stride + i]);
1.1309 + }
1.1310 +}
1.1311 +
1.1312 +void vp9_idct32x32_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
1.1313 + int i, j;
1.1314 + int a1;
1.1315 +
1.1316 + int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
1.1317 + out = dct_const_round_shift(out * cospi_16_64);
1.1318 + a1 = ROUND_POWER_OF_TWO(out, 6);
1.1319 +
1.1320 + for (j = 0; j < 32; ++j) {
1.1321 + for (i = 0; i < 32; ++i)
1.1322 + dest[i] = clip_pixel(dest[i] + a1);
1.1323 + dest += stride;
1.1324 + }
1.1325 +}
1.1326 +
1.1327 +// idct
1.1328 +void vp9_idct4x4_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
1.1329 + if (eob > 1)
1.1330 + vp9_idct4x4_16_add(input, dest, stride);
1.1331 + else
1.1332 + vp9_idct4x4_1_add(input, dest, stride);
1.1333 +}
1.1334 +
1.1335 +
1.1336 +void vp9_iwht4x4_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
1.1337 + if (eob > 1)
1.1338 + vp9_iwht4x4_16_add(input, dest, stride);
1.1339 + else
1.1340 + vp9_iwht4x4_1_add(input, dest, stride);
1.1341 +}
1.1342 +
1.1343 +void vp9_idct8x8_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
1.1344 + // If dc is 1, then input[0] is the reconstructed value, do not need
1.1345 + // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
1.1346 +
1.1347 + // The calculation can be simplified if there are not many non-zero dct
1.1348 + // coefficients. Use eobs to decide what to do.
1.1349 + // TODO(yunqingwang): "eobs = 1" case is also handled in vp9_short_idct8x8_c.
1.1350 + // Combine that with code here.
1.1351 + if (eob) {
1.1352 + if (eob == 1)
1.1353 + // DC only DCT coefficient
1.1354 + vp9_idct8x8_1_add(input, dest, stride);
1.1355 + else if (eob <= 10)
1.1356 + vp9_idct8x8_10_add(input, dest, stride);
1.1357 + else
1.1358 + vp9_idct8x8_64_add(input, dest, stride);
1.1359 + }
1.1360 +}
1.1361 +
1.1362 +void vp9_idct16x16_add(const int16_t *input, uint8_t *dest, int stride,
1.1363 + int eob) {
1.1364 + /* The calculation can be simplified if there are not many non-zero dct
1.1365 + * coefficients. Use eobs to separate different cases. */
1.1366 + if (eob) {
1.1367 + if (eob == 1)
1.1368 + /* DC only DCT coefficient. */
1.1369 + vp9_idct16x16_1_add(input, dest, stride);
1.1370 + else if (eob <= 10)
1.1371 + vp9_idct16x16_10_add(input, dest, stride);
1.1372 + else
1.1373 + vp9_idct16x16_256_add(input, dest, stride);
1.1374 + }
1.1375 +}
1.1376 +
1.1377 +void vp9_idct32x32_add(const int16_t *input, uint8_t *dest, int stride,
1.1378 + int eob) {
1.1379 + if (eob) {
1.1380 + if (eob == 1)
1.1381 + vp9_idct32x32_1_add(input, dest, stride);
1.1382 + else if (eob <= 34)
1.1383 + // non-zero coeff only in upper-left 8x8
1.1384 + vp9_idct32x32_34_add(input, dest, stride);
1.1385 + else
1.1386 + vp9_idct32x32_1024_add(input, dest, stride);
1.1387 + }
1.1388 +}
1.1389 +
1.1390 +// iht
1.1391 +void vp9_iht4x4_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
1.1392 + int stride, int eob) {
1.1393 + if (tx_type == DCT_DCT)
1.1394 + vp9_idct4x4_add(input, dest, stride, eob);
1.1395 + else
1.1396 + vp9_iht4x4_16_add(input, dest, stride, tx_type);
1.1397 +}
1.1398 +
1.1399 +void vp9_iht8x8_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
1.1400 + int stride, int eob) {
1.1401 + if (tx_type == DCT_DCT) {
1.1402 + vp9_idct8x8_add(input, dest, stride, eob);
1.1403 + } else {
1.1404 + if (eob > 0) {
1.1405 + vp9_iht8x8_64_add(input, dest, stride, tx_type);
1.1406 + }
1.1407 + }
1.1408 +}
1.1409 +
1.1410 +void vp9_iht16x16_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
1.1411 + int stride, int eob) {
1.1412 + if (tx_type == DCT_DCT) {
1.1413 + vp9_idct16x16_add(input, dest, stride, eob);
1.1414 + } else {
1.1415 + if (eob > 0) {
1.1416 + vp9_iht16x16_256_add(input, dest, stride, tx_type);
1.1417 + }
1.1418 + }
1.1419 +}
