michael@0: /*
michael@0:  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
michael@0:  *
michael@0:  *  Use of this source code is governed by a BSD-style license
michael@0:  *  that can be found in the LICENSE file in the root of the source
michael@0:  *  tree. An additional intellectual property rights grant can be found
michael@0:  *  in the file PATENTS.  All contributing project authors may
michael@0:  *  be found in the AUTHORS file in the root of the source tree.
michael@0:  */
michael@0: 
michael@0: #include <assert.h>
michael@0: #include <math.h>
michael@0: 
michael@0: #include "./vpx_config.h"
michael@0: #include "./vp9_rtcd.h"
michael@0: #include "vp9/common/vp9_systemdependent.h"
michael@0: #include "vp9/common/vp9_blockd.h"
michael@0: #include "vp9/common/vp9_common.h"
michael@0: #include "vp9/common/vp9_idct.h"
michael@0: 
michael@0: void vp9_iwht4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride) {
michael@0: /* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
michael@0:    0.5 shifts per pixel. */
michael@0:   int i;
michael@0:   int16_t output[16];
michael@0:   int a1, b1, c1, d1, e1;
michael@0:   const int16_t *ip = input;
michael@0:   int16_t *op = output;
michael@0: 
michael@0:   for (i = 0; i < 4; i++) {
michael@0:     a1 = ip[0] >> UNIT_QUANT_SHIFT;
michael@0:     c1 = ip[1] >> UNIT_QUANT_SHIFT;
michael@0:     d1 = ip[2] >> UNIT_QUANT_SHIFT;
michael@0:     b1 = ip[3] >> UNIT_QUANT_SHIFT;
michael@0:     a1 += c1;
michael@0:     d1 -= b1;
michael@0:     e1 = (a1 - d1) >> 1;
michael@0:     b1 = e1 - b1;
michael@0:     c1 = e1 - c1;
michael@0:     a1 -= b1;
michael@0:     d1 += c1;
michael@0:     op[0] = a1;
michael@0:     op[1] = b1;
michael@0:     op[2] = c1;
michael@0:     op[3] = d1;
michael@0:     ip += 4;
michael@0:     op += 4;
michael@0:   }
michael@0: 
michael@0:   ip = output;
michael@0:   for (i = 0; i < 4; i++) {
michael@0:     a1 = ip[4 * 0];
michael@0:     c1 = ip[4 * 1];
michael@0:     d1 = ip[4 * 2];
michael@0:     b1 = ip[4 * 3];
michael@0:     a1 += c1;
michael@0:     d1 -= b1;
michael@0:     e1 = (a1 - d1) >> 1;
michael@0:     b1 = e1 - b1;
michael@0:     c1 = e1 - c1;
michael@0:     a1 -= b1;
michael@0:     d1 += c1;
michael@0:     dest[stride * 0] = clip_pixel(dest[stride * 0] + a1);
michael@0:     dest[stride * 1] = clip_pixel(dest[stride * 1] + b1);
michael@0:     dest[stride * 2] = clip_pixel(dest[stride * 2] + c1);
michael@0:     dest[stride * 3] = clip_pixel(dest[stride * 3] + d1);
michael@0: 
michael@0:     ip++;
michael@0:     dest++;
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_iwht4x4_1_add_c(const int16_t *in, uint8_t *dest, int dest_stride) {
michael@0:   int i;
michael@0:   int a1, e1;
michael@0:   int16_t tmp[4];
michael@0:   const int16_t *ip = in;
michael@0:   int16_t *op = tmp;
michael@0: 
michael@0:   a1 = ip[0] >> UNIT_QUANT_SHIFT;
michael@0:   e1 = a1 >> 1;
michael@0:   a1 -= e1;
michael@0:   op[0] = a1;
michael@0:   op[1] = op[2] = op[3] = e1;
michael@0: 
michael@0:   ip = tmp;
michael@0:   for (i = 0; i < 4; i++) {
michael@0:     e1 = ip[0] >> 1;
michael@0:     a1 = ip[0] - e1;
michael@0:     dest[dest_stride * 0] = clip_pixel(dest[dest_stride * 0] + a1);
michael@0:     dest[dest_stride * 1] = clip_pixel(dest[dest_stride * 1] + e1);
michael@0:     dest[dest_stride * 2] = clip_pixel(dest[dest_stride * 2] + e1);
michael@0:     dest[dest_stride * 3] = clip_pixel(dest[dest_stride * 3] + e1);
michael@0:     ip++;
michael@0:     dest++;
michael@0:   }
michael@0: }
michael@0: 
michael@0: static void idct4_1d(const int16_t *input, int16_t *output) {
michael@0:   int16_t step[4];
michael@0:   int temp1, temp2;
michael@0:   // stage 1
michael@0:   temp1 = (input[0] + input[2]) * cospi_16_64;
michael@0:   temp2 = (input[0] - input[2]) * cospi_16_64;
michael@0:   step[0] = dct_const_round_shift(temp1);
michael@0:   step[1] = dct_const_round_shift(temp2);
michael@0:   temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64;
michael@0:   temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64;
michael@0:   step[2] = dct_const_round_shift(temp1);
michael@0:   step[3] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   // stage 2
michael@0:   output[0] = step[0] + step[3];
michael@0:   output[1] = step[1] + step[2];
michael@0:   output[2] = step[1] - step[2];
michael@0:   output[3] = step[0] - step[3];
michael@0: }
michael@0: 
michael@0: void vp9_idct4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride) {
michael@0:   int16_t out[4 * 4];
michael@0:   int16_t *outptr = out;
michael@0:   int i, j;
michael@0:   int16_t temp_in[4], temp_out[4];
michael@0: 
michael@0:   // Rows
michael@0:   for (i = 0; i < 4; ++i) {
michael@0:     idct4_1d(input, outptr);
michael@0:     input += 4;
michael@0:     outptr += 4;
michael@0:   }
michael@0: 
michael@0:   // Columns
michael@0:   for (i = 0; i < 4; ++i) {
michael@0:     for (j = 0; j < 4; ++j)
michael@0:       temp_in[j] = out[j * 4 + i];
michael@0:     idct4_1d(temp_in, temp_out);
michael@0:     for (j = 0; j < 4; ++j)
michael@0:       dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
michael@0:                                   + dest[j * stride + i]);
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_idct4x4_1_add_c(const int16_t *input, uint8_t *dest, int dest_stride) {
michael@0:   int i;
michael@0:   int a1;
michael@0:   int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
michael@0:   out = dct_const_round_shift(out * cospi_16_64);
michael@0:   a1 = ROUND_POWER_OF_TWO(out, 4);
michael@0: 
michael@0:   for (i = 0; i < 4; i++) {
michael@0:     dest[0] = clip_pixel(dest[0] + a1);
michael@0:     dest[1] = clip_pixel(dest[1] + a1);
michael@0:     dest[2] = clip_pixel(dest[2] + a1);
michael@0:     dest[3] = clip_pixel(dest[3] + a1);
michael@0:     dest += dest_stride;
michael@0:   }
michael@0: }
michael@0: 
michael@0: static void idct8_1d(const int16_t *input, int16_t *output) {
michael@0:   int16_t step1[8], step2[8];
michael@0:   int temp1, temp2;
michael@0:   // stage 1
michael@0:   step1[0] = input[0];
michael@0:   step1[2] = input[4];
michael@0:   step1[1] = input[2];
michael@0:   step1[3] = input[6];
michael@0:   temp1 = input[1] * cospi_28_64 - input[7] * cospi_4_64;
michael@0:   temp2 = input[1] * cospi_4_64 + input[7] * cospi_28_64;
michael@0:   step1[4] = dct_const_round_shift(temp1);
michael@0:   step1[7] = dct_const_round_shift(temp2);
michael@0:   temp1 = input[5] * cospi_12_64 - input[3] * cospi_20_64;
michael@0:   temp2 = input[5] * cospi_20_64 + input[3] * cospi_12_64;
michael@0:   step1[5] = dct_const_round_shift(temp1);
michael@0:   step1[6] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   // stage 2 & stage 3 - even half
michael@0:   idct4_1d(step1, step1);
michael@0: 
michael@0:   // stage 2 - odd half
michael@0:   step2[4] = step1[4] + step1[5];
michael@0:   step2[5] = step1[4] - step1[5];
michael@0:   step2[6] = -step1[6] + step1[7];
michael@0:   step2[7] = step1[6] + step1[7];
michael@0: 
michael@0:   // stage 3 -odd half
michael@0:   step1[4] = step2[4];
michael@0:   temp1 = (step2[6] - step2[5]) * cospi_16_64;
michael@0:   temp2 = (step2[5] + step2[6]) * cospi_16_64;
michael@0:   step1[5] = dct_const_round_shift(temp1);
michael@0:   step1[6] = dct_const_round_shift(temp2);
michael@0:   step1[7] = step2[7];
michael@0: 
michael@0:   // stage 4
michael@0:   output[0] = step1[0] + step1[7];
michael@0:   output[1] = step1[1] + step1[6];
michael@0:   output[2] = step1[2] + step1[5];
michael@0:   output[3] = step1[3] + step1[4];
michael@0:   output[4] = step1[3] - step1[4];
michael@0:   output[5] = step1[2] - step1[5];
michael@0:   output[6] = step1[1] - step1[6];
michael@0:   output[7] = step1[0] - step1[7];
michael@0: }
michael@0: 
michael@0: void vp9_idct8x8_64_add_c(const int16_t *input, uint8_t *dest, int stride) {
michael@0:   int16_t out[8 * 8];
michael@0:   int16_t *outptr = out;
michael@0:   int i, j;
michael@0:   int16_t temp_in[8], temp_out[8];
michael@0: 
michael@0:   // First transform rows
michael@0:   for (i = 0; i < 8; ++i) {
michael@0:     idct8_1d(input, outptr);
michael@0:     input += 8;
michael@0:     outptr += 8;
michael@0:   }
michael@0: 
michael@0:   // Then transform columns
michael@0:   for (i = 0; i < 8; ++i) {
michael@0:     for (j = 0; j < 8; ++j)
michael@0:       temp_in[j] = out[j * 8 + i];
michael@0:     idct8_1d(temp_in, temp_out);
michael@0:     for (j = 0; j < 8; ++j)
michael@0:       dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
michael@0:                                   + dest[j * stride + i]);
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_idct8x8_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
michael@0:   int i, j;
michael@0:   int a1;
michael@0:   int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
michael@0:   out = dct_const_round_shift(out * cospi_16_64);
michael@0:   a1 = ROUND_POWER_OF_TWO(out, 5);
michael@0:   for (j = 0; j < 8; ++j) {
michael@0:     for (i = 0; i < 8; ++i)
michael@0:       dest[i] = clip_pixel(dest[i] + a1);
michael@0:     dest += stride;
michael@0:   }
michael@0: }
michael@0: 
michael@0: static void iadst4_1d(const int16_t *input, int16_t *output) {
michael@0:   int s0, s1, s2, s3, s4, s5, s6, s7;
michael@0: 
michael@0:   int x0 = input[0];
michael@0:   int x1 = input[1];
michael@0:   int x2 = input[2];
michael@0:   int x3 = input[3];
michael@0: 
michael@0:   if (!(x0 | x1 | x2 | x3)) {
michael@0:     output[0] = output[1] = output[2] = output[3] = 0;
michael@0:     return;
michael@0:   }
michael@0: 
michael@0:   s0 = sinpi_1_9 * x0;
michael@0:   s1 = sinpi_2_9 * x0;
michael@0:   s2 = sinpi_3_9 * x1;
michael@0:   s3 = sinpi_4_9 * x2;
michael@0:   s4 = sinpi_1_9 * x2;
michael@0:   s5 = sinpi_2_9 * x3;
michael@0:   s6 = sinpi_4_9 * x3;
michael@0:   s7 = x0 - x2 + x3;
michael@0: 
michael@0:   x0 = s0 + s3 + s5;
michael@0:   x1 = s1 - s4 - s6;
michael@0:   x2 = sinpi_3_9 * s7;
michael@0:   x3 = s2;
michael@0: 
michael@0:   s0 = x0 + x3;
michael@0:   s1 = x1 + x3;
michael@0:   s2 = x2;
michael@0:   s3 = x0 + x1 - x3;
michael@0: 
michael@0:   // 1-D transform scaling factor is sqrt(2).
michael@0:   // The overall dynamic range is 14b (input) + 14b (multiplication scaling)
michael@0:   // + 1b (addition) = 29b.
michael@0:   // Hence the output bit depth is 15b.
michael@0:   output[0] = dct_const_round_shift(s0);
michael@0:   output[1] = dct_const_round_shift(s1);
michael@0:   output[2] = dct_const_round_shift(s2);
michael@0:   output[3] = dct_const_round_shift(s3);
michael@0: }
michael@0: 
michael@0: void vp9_iht4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride,
michael@0:                          int tx_type) {
michael@0:   const transform_2d IHT_4[] = {
michael@0:     { idct4_1d, idct4_1d  },  // DCT_DCT  = 0
michael@0:     { iadst4_1d, idct4_1d  },   // ADST_DCT = 1
michael@0:     { idct4_1d, iadst4_1d },    // DCT_ADST = 2
michael@0:     { iadst4_1d, iadst4_1d }      // ADST_ADST = 3
michael@0:   };
michael@0: 
michael@0:   int i, j;
michael@0:   int16_t out[4 * 4];
michael@0:   int16_t *outptr = out;
michael@0:   int16_t temp_in[4], temp_out[4];
michael@0: 
michael@0:   // inverse transform row vectors
michael@0:   for (i = 0; i < 4; ++i) {
michael@0:     IHT_4[tx_type].rows(input, outptr);
michael@0:     input  += 4;
michael@0:     outptr += 4;
michael@0:   }
michael@0: 
michael@0:   // inverse transform column vectors
michael@0:   for (i = 0; i < 4; ++i) {
michael@0:     for (j = 0; j < 4; ++j)
michael@0:       temp_in[j] = out[j * 4 + i];
michael@0:     IHT_4[tx_type].cols(temp_in, temp_out);
michael@0:     for (j = 0; j < 4; ++j)
michael@0:       dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
michael@0:                                   + dest[j * stride + i]);
michael@0:   }
michael@0: }
michael@0: static void iadst8_1d(const int16_t *input, int16_t *output) {
michael@0:   int s0, s1, s2, s3, s4, s5, s6, s7;
michael@0: 
michael@0:   int x0 = input[7];
michael@0:   int x1 = input[0];
michael@0:   int x2 = input[5];
michael@0:   int x3 = input[2];
michael@0:   int x4 = input[3];
michael@0:   int x5 = input[4];
michael@0:   int x6 = input[1];
michael@0:   int x7 = input[6];
michael@0: 
michael@0:   if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7)) {
michael@0:     output[0] = output[1] = output[2] = output[3] = output[4]
michael@0:               = output[5] = output[6] = output[7] = 0;
michael@0:     return;
michael@0:   }
michael@0: 
michael@0:   // stage 1
michael@0:   s0 = cospi_2_64  * x0 + cospi_30_64 * x1;
michael@0:   s1 = cospi_30_64 * x0 - cospi_2_64  * x1;
michael@0:   s2 = cospi_10_64 * x2 + cospi_22_64 * x3;
michael@0:   s3 = cospi_22_64 * x2 - cospi_10_64 * x3;
michael@0:   s4 = cospi_18_64 * x4 + cospi_14_64 * x5;
michael@0:   s5 = cospi_14_64 * x4 - cospi_18_64 * x5;
michael@0:   s6 = cospi_26_64 * x6 + cospi_6_64  * x7;
michael@0:   s7 = cospi_6_64  * x6 - cospi_26_64 * x7;
michael@0: 
michael@0:   x0 = dct_const_round_shift(s0 + s4);
michael@0:   x1 = dct_const_round_shift(s1 + s5);
michael@0:   x2 = dct_const_round_shift(s2 + s6);
michael@0:   x3 = dct_const_round_shift(s3 + s7);
michael@0:   x4 = dct_const_round_shift(s0 - s4);
michael@0:   x5 = dct_const_round_shift(s1 - s5);
michael@0:   x6 = dct_const_round_shift(s2 - s6);
michael@0:   x7 = dct_const_round_shift(s3 - s7);
michael@0: 
michael@0:   // stage 2
michael@0:   s0 = x0;
michael@0:   s1 = x1;
michael@0:   s2 = x2;
michael@0:   s3 = x3;
michael@0:   s4 =  cospi_8_64  * x4 + cospi_24_64 * x5;
michael@0:   s5 =  cospi_24_64 * x4 - cospi_8_64  * x5;
michael@0:   s6 = -cospi_24_64 * x6 + cospi_8_64  * x7;
michael@0:   s7 =  cospi_8_64  * x6 + cospi_24_64 * x7;
michael@0: 
michael@0:   x0 = s0 + s2;
michael@0:   x1 = s1 + s3;
michael@0:   x2 = s0 - s2;
michael@0:   x3 = s1 - s3;
michael@0:   x4 = dct_const_round_shift(s4 + s6);
michael@0:   x5 = dct_const_round_shift(s5 + s7);
michael@0:   x6 = dct_const_round_shift(s4 - s6);
michael@0:   x7 = dct_const_round_shift(s5 - s7);
michael@0: 
michael@0:   // stage 3
michael@0:   s2 = cospi_16_64 * (x2 + x3);
michael@0:   s3 = cospi_16_64 * (x2 - x3);
michael@0:   s6 = cospi_16_64 * (x6 + x7);
michael@0:   s7 = cospi_16_64 * (x6 - x7);
michael@0: 
michael@0:   x2 = dct_const_round_shift(s2);
michael@0:   x3 = dct_const_round_shift(s3);
michael@0:   x6 = dct_const_round_shift(s6);
michael@0:   x7 = dct_const_round_shift(s7);
michael@0: 
michael@0:   output[0] =  x0;
michael@0:   output[1] = -x4;
michael@0:   output[2] =  x6;
michael@0:   output[3] = -x2;
michael@0:   output[4] =  x3;
michael@0:   output[5] = -x7;
michael@0:   output[6] =  x5;
michael@0:   output[7] = -x1;
michael@0: }
michael@0: 
michael@0: static const transform_2d IHT_8[] = {
michael@0:   { idct8_1d,  idct8_1d  },  // DCT_DCT  = 0
michael@0:   { iadst8_1d, idct8_1d  },  // ADST_DCT = 1
michael@0:   { idct8_1d,  iadst8_1d },  // DCT_ADST = 2
michael@0:   { iadst8_1d, iadst8_1d }   // ADST_ADST = 3
michael@0: };
michael@0: 
michael@0: void vp9_iht8x8_64_add_c(const int16_t *input, uint8_t *dest, int stride,
michael@0:                          int tx_type) {
michael@0:   int i, j;
michael@0:   int16_t out[8 * 8];
michael@0:   int16_t *outptr = out;
michael@0:   int16_t temp_in[8], temp_out[8];
michael@0:   const transform_2d ht = IHT_8[tx_type];
michael@0: 
michael@0:   // inverse transform row vectors
michael@0:   for (i = 0; i < 8; ++i) {
michael@0:     ht.rows(input, outptr);
michael@0:     input += 8;
michael@0:     outptr += 8;
michael@0:   }
michael@0: 
michael@0:   // inverse transform column vectors
michael@0:   for (i = 0; i < 8; ++i) {
michael@0:     for (j = 0; j < 8; ++j)
michael@0:       temp_in[j] = out[j * 8 + i];
michael@0:     ht.cols(temp_in, temp_out);
michael@0:     for (j = 0; j < 8; ++j)
michael@0:       dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
michael@0:                                   + dest[j * stride + i]);
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_idct8x8_10_add_c(const int16_t *input, uint8_t *dest, int stride) {
michael@0:   int16_t out[8 * 8] = { 0 };
michael@0:   int16_t *outptr = out;
michael@0:   int i, j;
michael@0:   int16_t temp_in[8], temp_out[8];
michael@0: 
michael@0:   // First transform rows
michael@0:   // only first 4 row has non-zero coefs
michael@0:   for (i = 0; i < 4; ++i) {
michael@0:     idct8_1d(input, outptr);
michael@0:     input += 8;
michael@0:     outptr += 8;
michael@0:   }
michael@0: 
michael@0:   // Then transform columns
michael@0:   for (i = 0; i < 8; ++i) {
michael@0:     for (j = 0; j < 8; ++j)
michael@0:       temp_in[j] = out[j * 8 + i];
michael@0:     idct8_1d(temp_in, temp_out);
michael@0:     for (j = 0; j < 8; ++j)
michael@0:       dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
michael@0:                                   + dest[j * stride + i]);
michael@0:   }
michael@0: }
michael@0: 
michael@0: static void idct16_1d(const int16_t *input, int16_t *output) {
michael@0:   int16_t step1[16], step2[16];
michael@0:   int temp1, temp2;
michael@0: 
michael@0:   // stage 1
michael@0:   step1[0] = input[0/2];
michael@0:   step1[1] = input[16/2];
michael@0:   step1[2] = input[8/2];
michael@0:   step1[3] = input[24/2];
michael@0:   step1[4] = input[4/2];
michael@0:   step1[5] = input[20/2];
michael@0:   step1[6] = input[12/2];
michael@0:   step1[7] = input[28/2];
michael@0:   step1[8] = input[2/2];
michael@0:   step1[9] = input[18/2];
michael@0:   step1[10] = input[10/2];
michael@0:   step1[11] = input[26/2];
michael@0:   step1[12] = input[6/2];
michael@0:   step1[13] = input[22/2];
michael@0:   step1[14] = input[14/2];
michael@0:   step1[15] = input[30/2];
michael@0: 
michael@0:   // stage 2
michael@0:   step2[0] = step1[0];
michael@0:   step2[1] = step1[1];
michael@0:   step2[2] = step1[2];
michael@0:   step2[3] = step1[3];
michael@0:   step2[4] = step1[4];
michael@0:   step2[5] = step1[5];
michael@0:   step2[6] = step1[6];
michael@0:   step2[7] = step1[7];
michael@0: 
michael@0:   temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64;
michael@0:   temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64;
michael@0:   step2[8] = dct_const_round_shift(temp1);
michael@0:   step2[15] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64;
michael@0:   temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64;
michael@0:   step2[9] = dct_const_round_shift(temp1);
michael@0:   step2[14] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64;
michael@0:   temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64;
michael@0:   step2[10] = dct_const_round_shift(temp1);
michael@0:   step2[13] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64;
michael@0:   temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64;
michael@0:   step2[11] = dct_const_round_shift(temp1);
michael@0:   step2[12] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   // stage 3
michael@0:   step1[0] = step2[0];
michael@0:   step1[1] = step2[1];
michael@0:   step1[2] = step2[2];
michael@0:   step1[3] = step2[3];
michael@0: 
michael@0:   temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64;
michael@0:   temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64;
michael@0:   step1[4] = dct_const_round_shift(temp1);
michael@0:   step1[7] = dct_const_round_shift(temp2);
michael@0:   temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64;
michael@0:   temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64;
michael@0:   step1[5] = dct_const_round_shift(temp1);
michael@0:   step1[6] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   step1[8] = step2[8] + step2[9];
michael@0:   step1[9] = step2[8] - step2[9];
michael@0:   step1[10] = -step2[10] + step2[11];
michael@0:   step1[11] = step2[10] + step2[11];
michael@0:   step1[12] = step2[12] + step2[13];
michael@0:   step1[13] = step2[12] - step2[13];
michael@0:   step1[14] = -step2[14] + step2[15];
michael@0:   step1[15] = step2[14] + step2[15];
michael@0: 
michael@0:   // stage 4
michael@0:   temp1 = (step1[0] + step1[1]) * cospi_16_64;
michael@0:   temp2 = (step1[0] - step1[1]) * cospi_16_64;
michael@0:   step2[0] = dct_const_round_shift(temp1);
michael@0:   step2[1] = dct_const_round_shift(temp2);
michael@0:   temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64;
michael@0:   temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64;
michael@0:   step2[2] = dct_const_round_shift(temp1);
michael@0:   step2[3] = dct_const_round_shift(temp2);
michael@0:   step2[4] = step1[4] + step1[5];
michael@0:   step2[5] = step1[4] - step1[5];
michael@0:   step2[6] = -step1[6] + step1[7];
michael@0:   step2[7] = step1[6] + step1[7];
michael@0: 
michael@0:   step2[8] = step1[8];
michael@0:   step2[15] = step1[15];
michael@0:   temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64;
michael@0:   temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64;
michael@0:   step2[9] = dct_const_round_shift(temp1);
michael@0:   step2[14] = dct_const_round_shift(temp2);
michael@0:   temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64;
michael@0:   temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64;
michael@0:   step2[10] = dct_const_round_shift(temp1);
michael@0:   step2[13] = dct_const_round_shift(temp2);
michael@0:   step2[11] = step1[11];
michael@0:   step2[12] = step1[12];
michael@0: 
michael@0:   // stage 5
michael@0:   step1[0] = step2[0] + step2[3];
michael@0:   step1[1] = step2[1] + step2[2];
michael@0:   step1[2] = step2[1] - step2[2];
michael@0:   step1[3] = step2[0] - step2[3];
michael@0:   step1[4] = step2[4];
michael@0:   temp1 = (step2[6] - step2[5]) * cospi_16_64;
michael@0:   temp2 = (step2[5] + step2[6]) * cospi_16_64;
michael@0:   step1[5] = dct_const_round_shift(temp1);
michael@0:   step1[6] = dct_const_round_shift(temp2);
michael@0:   step1[7] = step2[7];
michael@0: 
michael@0:   step1[8] = step2[8] + step2[11];
michael@0:   step1[9] = step2[9] + step2[10];
michael@0:   step1[10] = step2[9] - step2[10];
michael@0:   step1[11] = step2[8] - step2[11];
michael@0:   step1[12] = -step2[12] + step2[15];
michael@0:   step1[13] = -step2[13] + step2[14];
michael@0:   step1[14] = step2[13] + step2[14];
michael@0:   step1[15] = step2[12] + step2[15];
michael@0: 
michael@0:   // stage 6
michael@0:   step2[0] = step1[0] + step1[7];
michael@0:   step2[1] = step1[1] + step1[6];
michael@0:   step2[2] = step1[2] + step1[5];
michael@0:   step2[3] = step1[3] + step1[4];
michael@0:   step2[4] = step1[3] - step1[4];
michael@0:   step2[5] = step1[2] - step1[5];
michael@0:   step2[6] = step1[1] - step1[6];
michael@0:   step2[7] = step1[0] - step1[7];
michael@0:   step2[8] = step1[8];
michael@0:   step2[9] = step1[9];
michael@0:   temp1 = (-step1[10] + step1[13]) * cospi_16_64;
michael@0:   temp2 = (step1[10] + step1[13]) * cospi_16_64;
michael@0:   step2[10] = dct_const_round_shift(temp1);
michael@0:   step2[13] = dct_const_round_shift(temp2);
michael@0:   temp1 = (-step1[11] + step1[12]) * cospi_16_64;
michael@0:   temp2 = (step1[11] + step1[12]) * cospi_16_64;
michael@0:   step2[11] = dct_const_round_shift(temp1);
michael@0:   step2[12] = dct_const_round_shift(temp2);
michael@0:   step2[14] = step1[14];
michael@0:   step2[15] = step1[15];
michael@0: 
michael@0:   // stage 7
michael@0:   output[0] = step2[0] + step2[15];
michael@0:   output[1] = step2[1] + step2[14];
michael@0:   output[2] = step2[2] + step2[13];
michael@0:   output[3] = step2[3] + step2[12];
michael@0:   output[4] = step2[4] + step2[11];
michael@0:   output[5] = step2[5] + step2[10];
michael@0:   output[6] = step2[6] + step2[9];
michael@0:   output[7] = step2[7] + step2[8];
michael@0:   output[8] = step2[7] - step2[8];
michael@0:   output[9] = step2[6] - step2[9];
michael@0:   output[10] = step2[5] - step2[10];
michael@0:   output[11] = step2[4] - step2[11];
michael@0:   output[12] = step2[3] - step2[12];
michael@0:   output[13] = step2[2] - step2[13];
michael@0:   output[14] = step2[1] - step2[14];
michael@0:   output[15] = step2[0] - step2[15];
michael@0: }
michael@0: 
michael@0: void vp9_idct16x16_256_add_c(const int16_t *input, uint8_t *dest, int stride) {
michael@0:   int16_t out[16 * 16];
michael@0:   int16_t *outptr = out;
michael@0:   int i, j;
michael@0:   int16_t temp_in[16], temp_out[16];
michael@0: 
michael@0:   // First transform rows
michael@0:   for (i = 0; i < 16; ++i) {
michael@0:     idct16_1d(input, outptr);
michael@0:     input += 16;
michael@0:     outptr += 16;
michael@0:   }
michael@0: 
michael@0:   // Then transform columns
michael@0:   for (i = 0; i < 16; ++i) {
michael@0:     for (j = 0; j < 16; ++j)
michael@0:       temp_in[j] = out[j * 16 + i];
michael@0:     idct16_1d(temp_in, temp_out);
michael@0:     for (j = 0; j < 16; ++j)
michael@0:       dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
michael@0:                                   + dest[j * stride + i]);
michael@0:   }
michael@0: }
michael@0: 
michael@0: static void iadst16_1d(const int16_t *input, int16_t *output) {
michael@0:   int s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15;
michael@0: 
michael@0:   int x0 = input[15];
michael@0:   int x1 = input[0];
michael@0:   int x2 = input[13];
michael@0:   int x3 = input[2];
michael@0:   int x4 = input[11];
michael@0:   int x5 = input[4];
michael@0:   int x6 = input[9];
michael@0:   int x7 = input[6];
michael@0:   int x8 = input[7];
michael@0:   int x9 = input[8];
michael@0:   int x10 = input[5];
michael@0:   int x11 = input[10];
michael@0:   int x12 = input[3];
michael@0:   int x13 = input[12];
michael@0:   int x14 = input[1];
michael@0:   int x15 = input[14];
michael@0: 
michael@0:   if (!(x0 | x1 | x2 | x3 | x4 | x5 | x6 | x7 | x8
michael@0:            | x9 | x10 | x11 | x12 | x13 | x14 | x15)) {
michael@0:     output[0] = output[1] = output[2] = output[3] = output[4]
michael@0:               = output[5] = output[6] = output[7] = output[8]
michael@0:               = output[9] = output[10] = output[11] = output[12]
michael@0:               = output[13] = output[14] = output[15] = 0;
michael@0:     return;
michael@0:   }
michael@0: 
michael@0:   // stage 1
michael@0:   s0 = x0 * cospi_1_64  + x1 * cospi_31_64;
michael@0:   s1 = x0 * cospi_31_64 - x1 * cospi_1_64;
michael@0:   s2 = x2 * cospi_5_64  + x3 * cospi_27_64;
michael@0:   s3 = x2 * cospi_27_64 - x3 * cospi_5_64;
michael@0:   s4 = x4 * cospi_9_64  + x5 * cospi_23_64;
michael@0:   s5 = x4 * cospi_23_64 - x5 * cospi_9_64;
michael@0:   s6 = x6 * cospi_13_64 + x7 * cospi_19_64;
michael@0:   s7 = x6 * cospi_19_64 - x7 * cospi_13_64;
michael@0:   s8 = x8 * cospi_17_64 + x9 * cospi_15_64;
michael@0:   s9 = x8 * cospi_15_64 - x9 * cospi_17_64;
michael@0:   s10 = x10 * cospi_21_64 + x11 * cospi_11_64;
michael@0:   s11 = x10 * cospi_11_64 - x11 * cospi_21_64;
michael@0:   s12 = x12 * cospi_25_64 + x13 * cospi_7_64;
michael@0:   s13 = x12 * cospi_7_64  - x13 * cospi_25_64;
michael@0:   s14 = x14 * cospi_29_64 + x15 * cospi_3_64;
michael@0:   s15 = x14 * cospi_3_64  - x15 * cospi_29_64;
michael@0: 
michael@0:   x0 = dct_const_round_shift(s0 + s8);
michael@0:   x1 = dct_const_round_shift(s1 + s9);
michael@0:   x2 = dct_const_round_shift(s2 + s10);
michael@0:   x3 = dct_const_round_shift(s3 + s11);
michael@0:   x4 = dct_const_round_shift(s4 + s12);
michael@0:   x5 = dct_const_round_shift(s5 + s13);
michael@0:   x6 = dct_const_round_shift(s6 + s14);
michael@0:   x7 = dct_const_round_shift(s7 + s15);
michael@0:   x8  = dct_const_round_shift(s0 - s8);
michael@0:   x9  = dct_const_round_shift(s1 - s9);
michael@0:   x10 = dct_const_round_shift(s2 - s10);
michael@0:   x11 = dct_const_round_shift(s3 - s11);
michael@0:   x12 = dct_const_round_shift(s4 - s12);
michael@0:   x13 = dct_const_round_shift(s5 - s13);
michael@0:   x14 = dct_const_round_shift(s6 - s14);
michael@0:   x15 = dct_const_round_shift(s7 - s15);
michael@0: 
michael@0:   // stage 2
michael@0:   s0 = x0;
michael@0:   s1 = x1;
michael@0:   s2 = x2;
michael@0:   s3 = x3;
michael@0:   s4 = x4;
michael@0:   s5 = x5;
michael@0:   s6 = x6;
michael@0:   s7 = x7;
michael@0:   s8 =    x8 * cospi_4_64   + x9 * cospi_28_64;
michael@0:   s9 =    x8 * cospi_28_64  - x9 * cospi_4_64;
michael@0:   s10 =   x10 * cospi_20_64 + x11 * cospi_12_64;
michael@0:   s11 =   x10 * cospi_12_64 - x11 * cospi_20_64;
michael@0:   s12 = - x12 * cospi_28_64 + x13 * cospi_4_64;
michael@0:   s13 =   x12 * cospi_4_64  + x13 * cospi_28_64;
michael@0:   s14 = - x14 * cospi_12_64 + x15 * cospi_20_64;
michael@0:   s15 =   x14 * cospi_20_64 + x15 * cospi_12_64;
michael@0: 
michael@0:   x0 = s0 + s4;
michael@0:   x1 = s1 + s5;
michael@0:   x2 = s2 + s6;
michael@0:   x3 = s3 + s7;
michael@0:   x4 = s0 - s4;
michael@0:   x5 = s1 - s5;
michael@0:   x6 = s2 - s6;
michael@0:   x7 = s3 - s7;
michael@0:   x8 = dct_const_round_shift(s8 + s12);
michael@0:   x9 = dct_const_round_shift(s9 + s13);
michael@0:   x10 = dct_const_round_shift(s10 + s14);
michael@0:   x11 = dct_const_round_shift(s11 + s15);
michael@0:   x12 = dct_const_round_shift(s8 - s12);
michael@0:   x13 = dct_const_round_shift(s9 - s13);
michael@0:   x14 = dct_const_round_shift(s10 - s14);
michael@0:   x15 = dct_const_round_shift(s11 - s15);
michael@0: 
michael@0:   // stage 3
michael@0:   s0 = x0;
michael@0:   s1 = x1;
michael@0:   s2 = x2;
michael@0:   s3 = x3;
michael@0:   s4 = x4 * cospi_8_64  + x5 * cospi_24_64;
michael@0:   s5 = x4 * cospi_24_64 - x5 * cospi_8_64;
michael@0:   s6 = - x6 * cospi_24_64 + x7 * cospi_8_64;
michael@0:   s7 =   x6 * cospi_8_64  + x7 * cospi_24_64;
michael@0:   s8 = x8;
michael@0:   s9 = x9;
michael@0:   s10 = x10;
michael@0:   s11 = x11;
michael@0:   s12 = x12 * cospi_8_64  + x13 * cospi_24_64;
michael@0:   s13 = x12 * cospi_24_64 - x13 * cospi_8_64;
michael@0:   s14 = - x14 * cospi_24_64 + x15 * cospi_8_64;
michael@0:   s15 =   x14 * cospi_8_64  + x15 * cospi_24_64;
michael@0: 
michael@0:   x0 = s0 + s2;
michael@0:   x1 = s1 + s3;
michael@0:   x2 = s0 - s2;
michael@0:   x3 = s1 - s3;
michael@0:   x4 = dct_const_round_shift(s4 + s6);
michael@0:   x5 = dct_const_round_shift(s5 + s7);
michael@0:   x6 = dct_const_round_shift(s4 - s6);
michael@0:   x7 = dct_const_round_shift(s5 - s7);
michael@0:   x8 = s8 + s10;
michael@0:   x9 = s9 + s11;
michael@0:   x10 = s8 - s10;
michael@0:   x11 = s9 - s11;
michael@0:   x12 = dct_const_round_shift(s12 + s14);
michael@0:   x13 = dct_const_round_shift(s13 + s15);
michael@0:   x14 = dct_const_round_shift(s12 - s14);
michael@0:   x15 = dct_const_round_shift(s13 - s15);
michael@0: 
michael@0:   // stage 4
michael@0:   s2 = (- cospi_16_64) * (x2 + x3);
michael@0:   s3 = cospi_16_64 * (x2 - x3);
michael@0:   s6 = cospi_16_64 * (x6 + x7);
michael@0:   s7 = cospi_16_64 * (- x6 + x7);
michael@0:   s10 = cospi_16_64 * (x10 + x11);
michael@0:   s11 = cospi_16_64 * (- x10 + x11);
michael@0:   s14 = (- cospi_16_64) * (x14 + x15);
michael@0:   s15 = cospi_16_64 * (x14 - x15);
michael@0: 
michael@0:   x2 = dct_const_round_shift(s2);
michael@0:   x3 = dct_const_round_shift(s3);
michael@0:   x6 = dct_const_round_shift(s6);
michael@0:   x7 = dct_const_round_shift(s7);
michael@0:   x10 = dct_const_round_shift(s10);
michael@0:   x11 = dct_const_round_shift(s11);
michael@0:   x14 = dct_const_round_shift(s14);
michael@0:   x15 = dct_const_round_shift(s15);
michael@0: 
michael@0:   output[0] =  x0;
michael@0:   output[1] = -x8;
michael@0:   output[2] =  x12;
michael@0:   output[3] = -x4;
michael@0:   output[4] =  x6;
michael@0:   output[5] =  x14;
michael@0:   output[6] =  x10;
michael@0:   output[7] =  x2;
michael@0:   output[8] =  x3;
michael@0:   output[9] =  x11;
michael@0:   output[10] =  x15;
michael@0:   output[11] =  x7;
michael@0:   output[12] =  x5;
michael@0:   output[13] = -x13;
michael@0:   output[14] =  x9;
michael@0:   output[15] = -x1;
michael@0: }
michael@0: 
michael@0: static const transform_2d IHT_16[] = {
michael@0:   { idct16_1d,  idct16_1d  },  // DCT_DCT  = 0
michael@0:   { iadst16_1d, idct16_1d  },  // ADST_DCT = 1
michael@0:   { idct16_1d,  iadst16_1d },  // DCT_ADST = 2
michael@0:   { iadst16_1d, iadst16_1d }   // ADST_ADST = 3
michael@0: };
michael@0: 
michael@0: void vp9_iht16x16_256_add_c(const int16_t *input, uint8_t *dest, int stride,
michael@0:                             int tx_type) {
michael@0:   int i, j;
michael@0:   int16_t out[16 * 16];
michael@0:   int16_t *outptr = out;
michael@0:   int16_t temp_in[16], temp_out[16];
michael@0:   const transform_2d ht = IHT_16[tx_type];
michael@0: 
michael@0:   // Rows
michael@0:   for (i = 0; i < 16; ++i) {
michael@0:     ht.rows(input, outptr);
michael@0:     input += 16;
michael@0:     outptr += 16;
michael@0:   }
michael@0: 
michael@0:   // Columns
michael@0:   for (i = 0; i < 16; ++i) {
michael@0:     for (j = 0; j < 16; ++j)
michael@0:       temp_in[j] = out[j * 16 + i];
michael@0:     ht.cols(temp_in, temp_out);
michael@0:     for (j = 0; j < 16; ++j)
michael@0:       dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
michael@0:                                         + dest[j * stride + i]);
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_idct16x16_10_add_c(const int16_t *input, uint8_t *dest, int stride) {
michael@0:   int16_t out[16 * 16] = { 0 };
michael@0:   int16_t *outptr = out;
michael@0:   int i, j;
michael@0:   int16_t temp_in[16], temp_out[16];
michael@0: 
michael@0:   // First transform rows. Since all non-zero dct coefficients are in
michael@0:   // upper-left 4x4 area, we only need to calculate first 4 rows here.
michael@0:   for (i = 0; i < 4; ++i) {
michael@0:     idct16_1d(input, outptr);
michael@0:     input += 16;
michael@0:     outptr += 16;
michael@0:   }
michael@0: 
michael@0:   // Then transform columns
michael@0:   for (i = 0; i < 16; ++i) {
michael@0:     for (j = 0; j < 16; ++j)
michael@0:       temp_in[j] = out[j*16 + i];
michael@0:     idct16_1d(temp_in, temp_out);
michael@0:     for (j = 0; j < 16; ++j)
michael@0:       dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
michael@0:                                   + dest[j * stride + i]);
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_idct16x16_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
michael@0:   int i, j;
michael@0:   int a1;
michael@0:   int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
michael@0:   out = dct_const_round_shift(out * cospi_16_64);
michael@0:   a1 = ROUND_POWER_OF_TWO(out, 6);
michael@0:   for (j = 0; j < 16; ++j) {
michael@0:     for (i = 0; i < 16; ++i)
michael@0:       dest[i] = clip_pixel(dest[i] + a1);
michael@0:     dest += stride;
michael@0:   }
michael@0: }
michael@0: 
michael@0: static void idct32_1d(const int16_t *input, int16_t *output) {
michael@0:   int16_t step1[32], step2[32];
michael@0:   int temp1, temp2;
michael@0: 
michael@0:   // stage 1
michael@0:   step1[0] = input[0];
michael@0:   step1[1] = input[16];
michael@0:   step1[2] = input[8];
michael@0:   step1[3] = input[24];
michael@0:   step1[4] = input[4];
michael@0:   step1[5] = input[20];
michael@0:   step1[6] = input[12];
michael@0:   step1[7] = input[28];
michael@0:   step1[8] = input[2];
michael@0:   step1[9] = input[18];
michael@0:   step1[10] = input[10];
michael@0:   step1[11] = input[26];
michael@0:   step1[12] = input[6];
michael@0:   step1[13] = input[22];
michael@0:   step1[14] = input[14];
michael@0:   step1[15] = input[30];
michael@0: 
michael@0:   temp1 = input[1] * cospi_31_64 - input[31] * cospi_1_64;
michael@0:   temp2 = input[1] * cospi_1_64 + input[31] * cospi_31_64;
michael@0:   step1[16] = dct_const_round_shift(temp1);
michael@0:   step1[31] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = input[17] * cospi_15_64 - input[15] * cospi_17_64;
michael@0:   temp2 = input[17] * cospi_17_64 + input[15] * cospi_15_64;
michael@0:   step1[17] = dct_const_round_shift(temp1);
michael@0:   step1[30] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = input[9] * cospi_23_64 - input[23] * cospi_9_64;
michael@0:   temp2 = input[9] * cospi_9_64 + input[23] * cospi_23_64;
michael@0:   step1[18] = dct_const_round_shift(temp1);
michael@0:   step1[29] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = input[25] * cospi_7_64 - input[7] * cospi_25_64;
michael@0:   temp2 = input[25] * cospi_25_64 + input[7] * cospi_7_64;
michael@0:   step1[19] = dct_const_round_shift(temp1);
michael@0:   step1[28] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = input[5] * cospi_27_64 - input[27] * cospi_5_64;
michael@0:   temp2 = input[5] * cospi_5_64 + input[27] * cospi_27_64;
michael@0:   step1[20] = dct_const_round_shift(temp1);
michael@0:   step1[27] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = input[21] * cospi_11_64 - input[11] * cospi_21_64;
michael@0:   temp2 = input[21] * cospi_21_64 + input[11] * cospi_11_64;
michael@0:   step1[21] = dct_const_round_shift(temp1);
michael@0:   step1[26] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = input[13] * cospi_19_64 - input[19] * cospi_13_64;
michael@0:   temp2 = input[13] * cospi_13_64 + input[19] * cospi_19_64;
michael@0:   step1[22] = dct_const_round_shift(temp1);
michael@0:   step1[25] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = input[29] * cospi_3_64 - input[3] * cospi_29_64;
michael@0:   temp2 = input[29] * cospi_29_64 + input[3] * cospi_3_64;
michael@0:   step1[23] = dct_const_round_shift(temp1);
michael@0:   step1[24] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   // stage 2
michael@0:   step2[0] = step1[0];
michael@0:   step2[1] = step1[1];
michael@0:   step2[2] = step1[2];
michael@0:   step2[3] = step1[3];
michael@0:   step2[4] = step1[4];
michael@0:   step2[5] = step1[5];
michael@0:   step2[6] = step1[6];
michael@0:   step2[7] = step1[7];
michael@0: 
michael@0:   temp1 = step1[8] * cospi_30_64 - step1[15] * cospi_2_64;
michael@0:   temp2 = step1[8] * cospi_2_64 + step1[15] * cospi_30_64;
michael@0:   step2[8] = dct_const_round_shift(temp1);
michael@0:   step2[15] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = step1[9] * cospi_14_64 - step1[14] * cospi_18_64;
michael@0:   temp2 = step1[9] * cospi_18_64 + step1[14] * cospi_14_64;
michael@0:   step2[9] = dct_const_round_shift(temp1);
michael@0:   step2[14] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = step1[10] * cospi_22_64 - step1[13] * cospi_10_64;
michael@0:   temp2 = step1[10] * cospi_10_64 + step1[13] * cospi_22_64;
michael@0:   step2[10] = dct_const_round_shift(temp1);
michael@0:   step2[13] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   temp1 = step1[11] * cospi_6_64 - step1[12] * cospi_26_64;
michael@0:   temp2 = step1[11] * cospi_26_64 + step1[12] * cospi_6_64;
michael@0:   step2[11] = dct_const_round_shift(temp1);
michael@0:   step2[12] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   step2[16] = step1[16] + step1[17];
michael@0:   step2[17] = step1[16] - step1[17];
michael@0:   step2[18] = -step1[18] + step1[19];
michael@0:   step2[19] = step1[18] + step1[19];
michael@0:   step2[20] = step1[20] + step1[21];
michael@0:   step2[21] = step1[20] - step1[21];
michael@0:   step2[22] = -step1[22] + step1[23];
michael@0:   step2[23] = step1[22] + step1[23];
michael@0:   step2[24] = step1[24] + step1[25];
michael@0:   step2[25] = step1[24] - step1[25];
michael@0:   step2[26] = -step1[26] + step1[27];
michael@0:   step2[27] = step1[26] + step1[27];
michael@0:   step2[28] = step1[28] + step1[29];
michael@0:   step2[29] = step1[28] - step1[29];
michael@0:   step2[30] = -step1[30] + step1[31];
michael@0:   step2[31] = step1[30] + step1[31];
michael@0: 
michael@0:   // stage 3
michael@0:   step1[0] = step2[0];
michael@0:   step1[1] = step2[1];
michael@0:   step1[2] = step2[2];
michael@0:   step1[3] = step2[3];
michael@0: 
michael@0:   temp1 = step2[4] * cospi_28_64 - step2[7] * cospi_4_64;
michael@0:   temp2 = step2[4] * cospi_4_64 + step2[7] * cospi_28_64;
michael@0:   step1[4] = dct_const_round_shift(temp1);
michael@0:   step1[7] = dct_const_round_shift(temp2);
michael@0:   temp1 = step2[5] * cospi_12_64 - step2[6] * cospi_20_64;
michael@0:   temp2 = step2[5] * cospi_20_64 + step2[6] * cospi_12_64;
michael@0:   step1[5] = dct_const_round_shift(temp1);
michael@0:   step1[6] = dct_const_round_shift(temp2);
michael@0: 
michael@0:   step1[8] = step2[8] + step2[9];
michael@0:   step1[9] = step2[8] - step2[9];
michael@0:   step1[10] = -step2[10] + step2[11];
michael@0:   step1[11] = step2[10] + step2[11];
michael@0:   step1[12] = step2[12] + step2[13];
michael@0:   step1[13] = step2[12] - step2[13];
michael@0:   step1[14] = -step2[14] + step2[15];
michael@0:   step1[15] = step2[14] + step2[15];
michael@0: 
michael@0:   step1[16] = step2[16];
michael@0:   step1[31] = step2[31];
michael@0:   temp1 = -step2[17] * cospi_4_64 + step2[30] * cospi_28_64;
michael@0:   temp2 = step2[17] * cospi_28_64 + step2[30] * cospi_4_64;
michael@0:   step1[17] = dct_const_round_shift(temp1);
michael@0:   step1[30] = dct_const_round_shift(temp2);
michael@0:   temp1 = -step2[18] * cospi_28_64 - step2[29] * cospi_4_64;
michael@0:   temp2 = -step2[18] * cospi_4_64 + step2[29] * cospi_28_64;
michael@0:   step1[18] = dct_const_round_shift(temp1);
michael@0:   step1[29] = dct_const_round_shift(temp2);
michael@0:   step1[19] = step2[19];
michael@0:   step1[20] = step2[20];
michael@0:   temp1 = -step2[21] * cospi_20_64 + step2[26] * cospi_12_64;
michael@0:   temp2 = step2[21] * cospi_12_64 + step2[26] * cospi_20_64;
michael@0:   step1[21] = dct_const_round_shift(temp1);
michael@0:   step1[26] = dct_const_round_shift(temp2);
michael@0:   temp1 = -step2[22] * cospi_12_64 - step2[25] * cospi_20_64;
michael@0:   temp2 = -step2[22] * cospi_20_64 + step2[25] * cospi_12_64;
michael@0:   step1[22] = dct_const_round_shift(temp1);
michael@0:   step1[25] = dct_const_round_shift(temp2);
michael@0:   step1[23] = step2[23];
michael@0:   step1[24] = step2[24];
michael@0:   step1[27] = step2[27];
michael@0:   step1[28] = step2[28];
michael@0: 
michael@0:   // stage 4
michael@0:   temp1 = (step1[0] + step1[1]) * cospi_16_64;
michael@0:   temp2 = (step1[0] - step1[1]) * cospi_16_64;
michael@0:   step2[0] = dct_const_round_shift(temp1);
michael@0:   step2[1] = dct_const_round_shift(temp2);
michael@0:   temp1 = step1[2] * cospi_24_64 - step1[3] * cospi_8_64;
michael@0:   temp2 = step1[2] * cospi_8_64 + step1[3] * cospi_24_64;
michael@0:   step2[2] = dct_const_round_shift(temp1);
michael@0:   step2[3] = dct_const_round_shift(temp2);
michael@0:   step2[4] = step1[4] + step1[5];
michael@0:   step2[5] = step1[4] - step1[5];
michael@0:   step2[6] = -step1[6] + step1[7];
michael@0:   step2[7] = step1[6] + step1[7];
michael@0: 
michael@0:   step2[8] = step1[8];
michael@0:   step2[15] = step1[15];
michael@0:   temp1 = -step1[9] * cospi_8_64 + step1[14] * cospi_24_64;
michael@0:   temp2 = step1[9] * cospi_24_64 + step1[14] * cospi_8_64;
michael@0:   step2[9] = dct_const_round_shift(temp1);
michael@0:   step2[14] = dct_const_round_shift(temp2);
michael@0:   temp1 = -step1[10] * cospi_24_64 - step1[13] * cospi_8_64;
michael@0:   temp2 = -step1[10] * cospi_8_64 + step1[13] * cospi_24_64;
michael@0:   step2[10] = dct_const_round_shift(temp1);
michael@0:   step2[13] = dct_const_round_shift(temp2);
michael@0:   step2[11] = step1[11];
michael@0:   step2[12] = step1[12];
michael@0: 
michael@0:   step2[16] = step1[16] + step1[19];
michael@0:   step2[17] = step1[17] + step1[18];
michael@0:   step2[18] = step1[17] - step1[18];
michael@0:   step2[19] = step1[16] - step1[19];
michael@0:   step2[20] = -step1[20] + step1[23];
michael@0:   step2[21] = -step1[21] + step1[22];
michael@0:   step2[22] = step1[21] + step1[22];
michael@0:   step2[23] = step1[20] + step1[23];
michael@0: 
michael@0:   step2[24] = step1[24] + step1[27];
michael@0:   step2[25] = step1[25] + step1[26];
michael@0:   step2[26] = step1[25] - step1[26];
michael@0:   step2[27] = step1[24] - step1[27];
michael@0:   step2[28] = -step1[28] + step1[31];
michael@0:   step2[29] = -step1[29] + step1[30];
michael@0:   step2[30] = step1[29] + step1[30];
michael@0:   step2[31] = step1[28] + step1[31];
michael@0: 
michael@0:   // stage 5
michael@0:   step1[0] = step2[0] + step2[3];
michael@0:   step1[1] = step2[1] + step2[2];
michael@0:   step1[2] = step2[1] - step2[2];
michael@0:   step1[3] = step2[0] - step2[3];
michael@0:   step1[4] = step2[4];
michael@0:   temp1 = (step2[6] - step2[5]) * cospi_16_64;
michael@0:   temp2 = (step2[5] + step2[6]) * cospi_16_64;
michael@0:   step1[5] = dct_const_round_shift(temp1);
michael@0:   step1[6] = dct_const_round_shift(temp2);
michael@0:   step1[7] = step2[7];
michael@0: 
michael@0:   step1[8] = step2[8] + step2[11];
michael@0:   step1[9] = step2[9] + step2[10];
michael@0:   step1[10] = step2[9] - step2[10];
michael@0:   step1[11] = step2[8] - step2[11];
michael@0:   step1[12] = -step2[12] + step2[15];
michael@0:   step1[13] = -step2[13] + step2[14];
michael@0:   step1[14] = step2[13] + step2[14];
michael@0:   step1[15] = step2[12] + step2[15];
michael@0: 
michael@0:   step1[16] = step2[16];
michael@0:   step1[17] = step2[17];
michael@0:   temp1 = -step2[18] * cospi_8_64 + step2[29] * cospi_24_64;
michael@0:   temp2 = step2[18] * cospi_24_64 + step2[29] * cospi_8_64;
michael@0:   step1[18] = dct_const_round_shift(temp1);
michael@0:   step1[29] = dct_const_round_shift(temp2);
michael@0:   temp1 = -step2[19] * cospi_8_64 + step2[28] * cospi_24_64;
michael@0:   temp2 = step2[19] * cospi_24_64 + step2[28] * cospi_8_64;
michael@0:   step1[19] = dct_const_round_shift(temp1);
michael@0:   step1[28] = dct_const_round_shift(temp2);
michael@0:   temp1 = -step2[20] * cospi_24_64 - step2[27] * cospi_8_64;
michael@0:   temp2 = -step2[20] * cospi_8_64 + step2[27] * cospi_24_64;
michael@0:   step1[20] = dct_const_round_shift(temp1);
michael@0:   step1[27] = dct_const_round_shift(temp2);
michael@0:   temp1 = -step2[21] * cospi_24_64 - step2[26] * cospi_8_64;
michael@0:   temp2 = -step2[21] * cospi_8_64 + step2[26] * cospi_24_64;
michael@0:   step1[21] = dct_const_round_shift(temp1);
michael@0:   step1[26] = dct_const_round_shift(temp2);
michael@0:   step1[22] = step2[22];
michael@0:   step1[23] = step2[23];
michael@0:   step1[24] = step2[24];
michael@0:   step1[25] = step2[25];
michael@0:   step1[30] = step2[30];
michael@0:   step1[31] = step2[31];
michael@0: 
michael@0:   // stage 6
michael@0:   step2[0] = step1[0] + step1[7];
michael@0:   step2[1] = step1[1] + step1[6];
michael@0:   step2[2] = step1[2] + step1[5];
michael@0:   step2[3] = step1[3] + step1[4];
michael@0:   step2[4] = step1[3] - step1[4];
michael@0:   step2[5] = step1[2] - step1[5];
michael@0:   step2[6] = step1[1] - step1[6];
michael@0:   step2[7] = step1[0] - step1[7];
michael@0:   step2[8] = step1[8];
michael@0:   step2[9] = step1[9];
michael@0:   temp1 = (-step1[10] + step1[13]) * cospi_16_64;
michael@0:   temp2 = (step1[10] + step1[13]) * cospi_16_64;
michael@0:   step2[10] = dct_const_round_shift(temp1);
michael@0:   step2[13] = dct_const_round_shift(temp2);
michael@0:   temp1 = (-step1[11] + step1[12]) * cospi_16_64;
michael@0:   temp2 = (step1[11] + step1[12]) * cospi_16_64;
michael@0:   step2[11] = dct_const_round_shift(temp1);
michael@0:   step2[12] = dct_const_round_shift(temp2);
michael@0:   step2[14] = step1[14];
michael@0:   step2[15] = step1[15];
michael@0: 
michael@0:   step2[16] = step1[16] + step1[23];
michael@0:   step2[17] = step1[17] + step1[22];
michael@0:   step2[18] = step1[18] + step1[21];
michael@0:   step2[19] = step1[19] + step1[20];
michael@0:   step2[20] = step1[19] - step1[20];
michael@0:   step2[21] = step1[18] - step1[21];
michael@0:   step2[22] = step1[17] - step1[22];
michael@0:   step2[23] = step1[16] - step1[23];
michael@0: 
michael@0:   step2[24] = -step1[24] + step1[31];
michael@0:   step2[25] = -step1[25] + step1[30];
michael@0:   step2[26] = -step1[26] + step1[29];
michael@0:   step2[27] = -step1[27] + step1[28];
michael@0:   step2[28] = step1[27] + step1[28];
michael@0:   step2[29] = step1[26] + step1[29];
michael@0:   step2[30] = step1[25] + step1[30];
michael@0:   step2[31] = step1[24] + step1[31];
michael@0: 
michael@0:   // stage 7
michael@0:   step1[0] = step2[0] + step2[15];
michael@0:   step1[1] = step2[1] + step2[14];
michael@0:   step1[2] = step2[2] + step2[13];
michael@0:   step1[3] = step2[3] + step2[12];
michael@0:   step1[4] = step2[4] + step2[11];
michael@0:   step1[5] = step2[5] + step2[10];
michael@0:   step1[6] = step2[6] + step2[9];
michael@0:   step1[7] = step2[7] + step2[8];
michael@0:   step1[8] = step2[7] - step2[8];
michael@0:   step1[9] = step2[6] - step2[9];
michael@0:   step1[10] = step2[5] - step2[10];
michael@0:   step1[11] = step2[4] - step2[11];
michael@0:   step1[12] = step2[3] - step2[12];
michael@0:   step1[13] = step2[2] - step2[13];
michael@0:   step1[14] = step2[1] - step2[14];
michael@0:   step1[15] = step2[0] - step2[15];
michael@0: 
michael@0:   step1[16] = step2[16];
michael@0:   step1[17] = step2[17];
michael@0:   step1[18] = step2[18];
michael@0:   step1[19] = step2[19];
michael@0:   temp1 = (-step2[20] + step2[27]) * cospi_16_64;
michael@0:   temp2 = (step2[20] + step2[27]) * cospi_16_64;
michael@0:   step1[20] = dct_const_round_shift(temp1);
michael@0:   step1[27] = dct_const_round_shift(temp2);
michael@0:   temp1 = (-step2[21] + step2[26]) * cospi_16_64;
michael@0:   temp2 = (step2[21] + step2[26]) * cospi_16_64;
michael@0:   step1[21] = dct_const_round_shift(temp1);
michael@0:   step1[26] = dct_const_round_shift(temp2);
michael@0:   temp1 = (-step2[22] + step2[25]) * cospi_16_64;
michael@0:   temp2 = (step2[22] + step2[25]) * cospi_16_64;
michael@0:   step1[22] = dct_const_round_shift(temp1);
michael@0:   step1[25] = dct_const_round_shift(temp2);
michael@0:   temp1 = (-step2[23] + step2[24]) * cospi_16_64;
michael@0:   temp2 = (step2[23] + step2[24]) * cospi_16_64;
michael@0:   step1[23] = dct_const_round_shift(temp1);
michael@0:   step1[24] = dct_const_round_shift(temp2);
michael@0:   step1[28] = step2[28];
michael@0:   step1[29] = step2[29];
michael@0:   step1[30] = step2[30];
michael@0:   step1[31] = step2[31];
michael@0: 
michael@0:   // final stage
michael@0:   output[0] = step1[0] + step1[31];
michael@0:   output[1] = step1[1] + step1[30];
michael@0:   output[2] = step1[2] + step1[29];
michael@0:   output[3] = step1[3] + step1[28];
michael@0:   output[4] = step1[4] + step1[27];
michael@0:   output[5] = step1[5] + step1[26];
michael@0:   output[6] = step1[6] + step1[25];
michael@0:   output[7] = step1[7] + step1[24];
michael@0:   output[8] = step1[8] + step1[23];
michael@0:   output[9] = step1[9] + step1[22];
michael@0:   output[10] = step1[10] + step1[21];
michael@0:   output[11] = step1[11] + step1[20];
michael@0:   output[12] = step1[12] + step1[19];
michael@0:   output[13] = step1[13] + step1[18];
michael@0:   output[14] = step1[14] + step1[17];
michael@0:   output[15] = step1[15] + step1[16];
michael@0:   output[16] = step1[15] - step1[16];
michael@0:   output[17] = step1[14] - step1[17];
michael@0:   output[18] = step1[13] - step1[18];
michael@0:   output[19] = step1[12] - step1[19];
michael@0:   output[20] = step1[11] - step1[20];
michael@0:   output[21] = step1[10] - step1[21];
michael@0:   output[22] = step1[9] - step1[22];
michael@0:   output[23] = step1[8] - step1[23];
michael@0:   output[24] = step1[7] - step1[24];
michael@0:   output[25] = step1[6] - step1[25];
michael@0:   output[26] = step1[5] - step1[26];
michael@0:   output[27] = step1[4] - step1[27];
michael@0:   output[28] = step1[3] - step1[28];
michael@0:   output[29] = step1[2] - step1[29];
michael@0:   output[30] = step1[1] - step1[30];
michael@0:   output[31] = step1[0] - step1[31];
michael@0: }
michael@0: 
michael@0: void vp9_idct32x32_1024_add_c(const int16_t *input, uint8_t *dest, int stride) {
michael@0:   int16_t out[32 * 32];
michael@0:   int16_t *outptr = out;
michael@0:   int i, j;
michael@0:   int16_t temp_in[32], temp_out[32];
michael@0: 
michael@0:   // Rows
michael@0:   for (i = 0; i < 32; ++i) {
michael@0:     int16_t zero_coeff[16];
michael@0:     for (j = 0; j < 16; ++j)
michael@0:       zero_coeff[j] = input[2 * j] | input[2 * j + 1];
michael@0:     for (j = 0; j < 8; ++j)
michael@0:       zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
michael@0:     for (j = 0; j < 4; ++j)
michael@0:       zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
michael@0:     for (j = 0; j < 2; ++j)
michael@0:       zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
michael@0: 
michael@0:     if (zero_coeff[0] | zero_coeff[1])
michael@0:       idct32_1d(input, outptr);
michael@0:     else
michael@0:       vpx_memset(outptr, 0, sizeof(int16_t) * 32);
michael@0:     input += 32;
michael@0:     outptr += 32;
michael@0:   }
michael@0: 
michael@0:   // Columns
michael@0:   for (i = 0; i < 32; ++i) {
michael@0:     for (j = 0; j < 32; ++j)
michael@0:       temp_in[j] = out[j * 32 + i];
michael@0:     idct32_1d(temp_in, temp_out);
michael@0:     for (j = 0; j < 32; ++j)
michael@0:       dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
michael@0:                                         + dest[j * stride + i]);
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_idct32x32_34_add_c(const int16_t *input, uint8_t *dest, int stride) {
michael@0:   int16_t out[32 * 32] = {0};
michael@0:   int16_t *outptr = out;
michael@0:   int i, j;
michael@0:   int16_t temp_in[32], temp_out[32];
michael@0: 
michael@0:   // Rows
michael@0:   // only upper-left 8x8 has non-zero coeff
michael@0:   for (i = 0; i < 8; ++i) {
michael@0:     idct32_1d(input, outptr);
michael@0:     input += 32;
michael@0:     outptr += 32;
michael@0:   }
michael@0: 
michael@0:   // Columns
michael@0:   for (i = 0; i < 32; ++i) {
michael@0:     for (j = 0; j < 32; ++j)
michael@0:       temp_in[j] = out[j * 32 + i];
michael@0:     idct32_1d(temp_in, temp_out);
michael@0:     for (j = 0; j < 32; ++j)
michael@0:       dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
michael@0:                                   + dest[j * stride + i]);
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_idct32x32_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
michael@0:   int i, j;
michael@0:   int a1;
michael@0: 
michael@0:   int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
michael@0:   out = dct_const_round_shift(out * cospi_16_64);
michael@0:   a1 = ROUND_POWER_OF_TWO(out, 6);
michael@0: 
michael@0:   for (j = 0; j < 32; ++j) {
michael@0:     for (i = 0; i < 32; ++i)
michael@0:       dest[i] = clip_pixel(dest[i] + a1);
michael@0:     dest += stride;
michael@0:   }
michael@0: }
michael@0: 
michael@0: // idct
michael@0: void vp9_idct4x4_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
michael@0:   if (eob > 1)
michael@0:     vp9_idct4x4_16_add(input, dest, stride);
michael@0:   else
michael@0:     vp9_idct4x4_1_add(input, dest, stride);
michael@0: }
michael@0: 
michael@0: 
michael@0: void vp9_iwht4x4_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
michael@0:   if (eob > 1)
michael@0:     vp9_iwht4x4_16_add(input, dest, stride);
michael@0:   else
michael@0:     vp9_iwht4x4_1_add(input, dest, stride);
michael@0: }
michael@0: 
michael@0: void vp9_idct8x8_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
michael@0:   // If dc is 1, then input[0] is the reconstructed value, do not need
michael@0:   // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
michael@0: 
michael@0:   // The calculation can be simplified if there are not many non-zero dct
michael@0:   // coefficients. Use eobs to decide what to do.
michael@0:   // TODO(yunqingwang): "eobs = 1" case is also handled in vp9_short_idct8x8_c.
michael@0:   // Combine that with code here.
michael@0:   if (eob) {
michael@0:     if (eob == 1)
michael@0:       // DC only DCT coefficient
michael@0:       vp9_idct8x8_1_add(input, dest, stride);
michael@0:     else if (eob <= 10)
michael@0:       vp9_idct8x8_10_add(input, dest, stride);
michael@0:     else
michael@0:       vp9_idct8x8_64_add(input, dest, stride);
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_idct16x16_add(const int16_t *input, uint8_t *dest, int stride,
michael@0:                        int eob) {
michael@0:   /* The calculation can be simplified if there are not many non-zero dct
michael@0:    * coefficients. Use eobs to separate different cases. */
michael@0:   if (eob) {
michael@0:     if (eob == 1)
michael@0:       /* DC only DCT coefficient. */
michael@0:       vp9_idct16x16_1_add(input, dest, stride);
michael@0:     else if (eob <= 10)
michael@0:       vp9_idct16x16_10_add(input, dest, stride);
michael@0:     else
michael@0:       vp9_idct16x16_256_add(input, dest, stride);
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_idct32x32_add(const int16_t *input, uint8_t *dest, int stride,
michael@0:                        int eob) {
michael@0:   if (eob) {
michael@0:     if (eob == 1)
michael@0:       vp9_idct32x32_1_add(input, dest, stride);
michael@0:     else if (eob <= 34)
michael@0:       // non-zero coeff only in upper-left 8x8
michael@0:       vp9_idct32x32_34_add(input, dest, stride);
michael@0:     else
michael@0:       vp9_idct32x32_1024_add(input, dest, stride);
michael@0:   }
michael@0: }
michael@0: 
michael@0: // iht
michael@0: void vp9_iht4x4_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
michael@0:                     int stride, int eob) {
michael@0:   if (tx_type == DCT_DCT)
michael@0:     vp9_idct4x4_add(input, dest, stride, eob);
michael@0:   else
michael@0:     vp9_iht4x4_16_add(input, dest, stride, tx_type);
michael@0: }
michael@0: 
michael@0: void vp9_iht8x8_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
michael@0:                     int stride, int eob) {
michael@0:   if (tx_type == DCT_DCT) {
michael@0:     vp9_idct8x8_add(input, dest, stride, eob);
michael@0:   } else {
michael@0:     if (eob > 0) {
michael@0:       vp9_iht8x8_64_add(input, dest, stride, tx_type);
michael@0:     }
michael@0:   }
michael@0: }
michael@0: 
michael@0: void vp9_iht16x16_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
michael@0:                       int stride, int eob) {
michael@0:   if (tx_type == DCT_DCT) {
michael@0:     vp9_idct16x16_add(input, dest, stride, eob);
michael@0:   } else {
michael@0:     if (eob > 0) {
michael@0:       vp9_iht16x16_256_add(input, dest, stride, tx_type);
michael@0:     }
michael@0:   }
michael@0: }