libvpx: Pull from upstream

source/libvpx/vp9/common/vp9_idct.c

 /*
  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
 #include <assert.h>
 #include <math.h>
 
 #include "./vpx_config.h"
 #include "./vp9_rtcd.h"
 #include "vp9/common/vp9_systemdependent.h"
 #include "vp9/common/vp9_blockd.h"
 #include "vp9/common/vp9_common.h"
 #include "vp9/common/vp9_idct.h"
 
-void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_iwht4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride) {
 /* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
    0.5 shifts per pixel. */
   int i;
   int16_t output[16];
   int a1, b1, c1, d1, e1;
-  int16_t *ip = input;
+  const int16_t *ip = input;
   int16_t *op = output;
 
   for (i = 0; i < 4; i++) {
-    a1 = ip[0] >> WHT_UPSCALE_FACTOR;
-    c1 = ip[1] >> WHT_UPSCALE_FACTOR;
-    d1 = ip[2] >> WHT_UPSCALE_FACTOR;
-    b1 = ip[3] >> WHT_UPSCALE_FACTOR;
+    a1 = ip[0] >> UNIT_QUANT_SHIFT;
+    c1 = ip[1] >> UNIT_QUANT_SHIFT;
+    d1 = ip[2] >> UNIT_QUANT_SHIFT;
+    b1 = ip[3] >> UNIT_QUANT_SHIFT;
     a1 += c1;
     d1 -= b1;
     e1 = (a1 - d1) >> 1;
     b1 = e1 - b1;
     c1 = e1 - c1;
     a1 -= b1;
     d1 += c1;
     op[0] = a1;
     op[1] = b1;
     op[2] = c1;
     op[3] = d1;
     ip += 4;
     op += 4;
   }
 
   ip = output;
   for (i = 0; i < 4; i++) {
     a1 = ip[4 * 0];
     c1 = ip[4 * 1];
     d1 = ip[4 * 2];
     b1 = ip[4 * 3];
     a1 += c1;
     d1 -= b1;
     e1 = (a1 - d1) >> 1;
     b1 = e1 - b1;
     c1 = e1 - c1;
     a1 -= b1;
     d1 += c1;
-    dest[dest_stride * 0] = clip_pixel(dest[dest_stride * 0] + a1);
-    dest[dest_stride * 1] = clip_pixel(dest[dest_stride * 1] + b1);
-    dest[dest_stride * 2] = clip_pixel(dest[dest_stride * 2] + c1);
-    dest[dest_stride * 3] = clip_pixel(dest[dest_stride * 3] + d1);
+    dest[stride * 0] = clip_pixel(dest[stride * 0] + a1);
+    dest[stride * 1] = clip_pixel(dest[stride * 1] + b1);
+    dest[stride * 2] = clip_pixel(dest[stride * 2] + c1);
+    dest[stride * 3] = clip_pixel(dest[stride * 3] + d1);
 
     ip++;
     dest++;
   }
 }
 
-void vp9_short_iwalsh4x4_1_add_c(int16_t *in, uint8_t *dest, int dest_stride) {
+void vp9_iwht4x4_1_add_c(const int16_t *in, uint8_t *dest, int dest_stride) {
   int i;
   int a1, e1;
   int16_t tmp[4];
-  int16_t *ip = in;
+  const int16_t *ip = in;
   int16_t *op = tmp;
 
-  a1 = ip[0] >> WHT_UPSCALE_FACTOR;
+  a1 = ip[0] >> UNIT_QUANT_SHIFT;
   e1 = a1 >> 1;
   a1 -= e1;
   op[0] = a1;
   op[1] = op[2] = op[3] = e1;
 
   ip = tmp;
   for (i = 0; i < 4; i++) {
     e1 = ip[0] >> 1;
     a1 = ip[0] - e1;
     dest[dest_stride * 0] = clip_pixel(dest[dest_stride * 0] + a1);
     dest[dest_stride * 1] = clip_pixel(dest[dest_stride * 1] + e1);
     dest[dest_stride * 2] = clip_pixel(dest[dest_stride * 2] + e1);
     dest[dest_stride * 3] = clip_pixel(dest[dest_stride * 3] + e1);
     ip++;
     dest++;
   }
 }
 
-void vp9_idct4_1d_c(int16_t *input, int16_t *output) {
+static void idct4_1d(const int16_t *input, int16_t *output) {
   int16_t step[4];
   int temp1, temp2;
   // stage 1
   temp1 = (input[0] + input[2]) * cospi_16_64;
   temp2 = (input[0] - input[2]) * cospi_16_64;
   step[0] = dct_const_round_shift(temp1);
   step[1] = dct_const_round_shift(temp2);
   temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64;
   temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64;
   step[2] = dct_const_round_shift(temp1);
   step[3] = dct_const_round_shift(temp2);
 
   // stage 2
   output[0] = step[0] + step[3];
   output[1] = step[1] + step[2];
   output[2] = step[1] - step[2];
   output[3] = step[0] - step[3];
 }
 
-void vp9_short_idct4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride) {
   int16_t out[4 * 4];
   int16_t *outptr = out;
   int i, j;
   int16_t temp_in[4], temp_out[4];
 
   // Rows
   for (i = 0; i < 4; ++i) {
-    vp9_idct4_1d(input, outptr);
+    idct4_1d(input, outptr);
     input += 4;
     outptr += 4;
   }
 
   // Columns
   for (i = 0; i < 4; ++i) {
     for (j = 0; j < 4; ++j)
       temp_in[j] = out[j * 4 + i];
-    vp9_idct4_1d(temp_in, temp_out);
+    idct4_1d(temp_in, temp_out);
     for (j = 0; j < 4; ++j)
-      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
-                                  + dest[j * dest_stride + i]);
+      dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
+                                  + dest[j * stride + i]);
   }
 }
 
-void vp9_short_idct4x4_1_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct4x4_1_add_c(const int16_t *input, uint8_t *dest, int dest_stride) {
   int i;
   int a1;
   int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
   out = dct_const_round_shift(out * cospi_16_64);
   a1 = ROUND_POWER_OF_TWO(out, 4);
 
   for (i = 0; i < 4; i++) {
     dest[0] = clip_pixel(dest[0] + a1);
     dest[1] = clip_pixel(dest[1] + a1);
     dest[2] = clip_pixel(dest[2] + a1);
     dest[3] = clip_pixel(dest[3] + a1);
     dest += dest_stride;
   }
 }
 
-static void idct8_1d(int16_t *input, int16_t *output) {
+static void idct8_1d(const int16_t *input, int16_t *output) {
   int16_t step1[8], step2[8];
   int temp1, temp2;
   // stage 1
   step1[0] = input[0];
   step1[2] = input[4];
   step1[1] = input[2];
   step1[3] = input[6];
   temp1 = input[1] * cospi_28_64 - input[7] * cospi_4_64;
   temp2 = input[1] * cospi_4_64 + input[7] * cospi_28_64;
   step1[4] = dct_const_round_shift(temp1);
   step1[7] = dct_const_round_shift(temp2);
   temp1 = input[5] * cospi_12_64 - input[3] * cospi_20_64;
   temp2 = input[5] * cospi_20_64 + input[3] * cospi_12_64;
   step1[5] = dct_const_round_shift(temp1);
   step1[6] = dct_const_round_shift(temp2);
 
   // stage 2 & stage 3 - even half
-  vp9_idct4_1d(step1, step1);
+  idct4_1d(step1, step1);
 
   // stage 2 - odd half
   step2[4] = step1[4] + step1[5];
   step2[5] = step1[4] - step1[5];
   step2[6] = -step1[6] + step1[7];
   step2[7] = step1[6] + step1[7];
 
   // stage 3 -odd half
   step1[4] = step2[4];
   temp1 = (step2[6] - step2[5]) * cospi_16_64;
   temp2 = (step2[5] + step2[6]) * cospi_16_64;
   step1[5] = dct_const_round_shift(temp1);
   step1[6] = dct_const_round_shift(temp2);
   step1[7] = step2[7];
 
   // stage 4
   output[0] = step1[0] + step1[7];
   output[1] = step1[1] + step1[6];
   output[2] = step1[2] + step1[5];
   output[3] = step1[3] + step1[4];
   output[4] = step1[3] - step1[4];
   output[5] = step1[2] - step1[5];
   output[6] = step1[1] - step1[6];
   output[7] = step1[0] - step1[7];
 }
 
-void vp9_short_idct8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct8x8_64_add_c(const int16_t *input, uint8_t *dest, int stride) {
   int16_t out[8 * 8];
   int16_t *outptr = out;
   int i, j;
   int16_t temp_in[8], temp_out[8];
 
   // First transform rows
   for (i = 0; i < 8; ++i) {
     idct8_1d(input, outptr);
     input += 8;
     outptr += 8;
   }
 
   // Then transform columns
   for (i = 0; i < 8; ++i) {
     for (j = 0; j < 8; ++j)
       temp_in[j] = out[j * 8 + i];
     idct8_1d(temp_in, temp_out);
     for (j = 0; j < 8; ++j)
-      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
-                                  + dest[j * dest_stride + i]);
+      dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
+                                  + dest[j * stride + i]);
   }
 }
 
-void vp9_short_idct8x8_1_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct8x8_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
   int i, j;
   int a1;
   int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
   out = dct_const_round_shift(out * cospi_16_64);
   a1 = ROUND_POWER_OF_TWO(out, 5);
   for (j = 0; j < 8; ++j) {
     for (i = 0; i < 8; ++i)
       dest[i] = clip_pixel(dest[i] + a1);
-    dest += dest_stride;
+    dest += stride;
   }
 }
 
-static void iadst4_1d(int16_t *input, int16_t *output) {
+static void iadst4_1d(const int16_t *input, int16_t *output) {
   int s0, s1, s2, s3, s4, s5, s6, s7;
 
   int x0 = input[0];
   int x1 = input[1];
   int x2 = input[2];
   int x3 = input[3];
 
   if (!(x0 | x1 | x2 | x3)) {
     output[0] = output[1] = output[2] = output[3] = 0;
     return;
@@ skipping 21 matching lines @@
   // 1-D transform scaling factor is sqrt(2).
   // The overall dynamic range is 14b (input) + 14b (multiplication scaling)
   // + 1b (addition) = 29b.
   // Hence the output bit depth is 15b.
   output[0] = dct_const_round_shift(s0);
   output[1] = dct_const_round_shift(s1);
   output[2] = dct_const_round_shift(s2);
   output[3] = dct_const_round_shift(s3);
 }
 
-void vp9_short_iht4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride,
-                            int tx_type) {
+void vp9_iht4x4_16_add_c(const int16_t *input, uint8_t *dest, int stride,
+                         int tx_type) {
   const transform_2d IHT_4[] = {
-    { vp9_idct4_1d,  vp9_idct4_1d  },  // DCT_DCT  = 0
-    { iadst4_1d, vp9_idct4_1d  },      // ADST_DCT = 1
-    { vp9_idct4_1d,  iadst4_1d },      // DCT_ADST = 2
-    { iadst4_1d, iadst4_1d }           // ADST_ADST = 3
+    { idct4_1d, idct4_1d  },  // DCT_DCT  = 0
+    { iadst4_1d, idct4_1d  },   // ADST_DCT = 1
+    { idct4_1d, iadst4_1d },    // DCT_ADST = 2
+    { iadst4_1d, iadst4_1d }      // ADST_ADST = 3
   };
 
   int i, j;
   int16_t out[4 * 4];
   int16_t *outptr = out;
   int16_t temp_in[4], temp_out[4];
 
   // inverse transform row vectors
   for (i = 0; i < 4; ++i) {
     IHT_4[tx_type].rows(input, outptr);
     input  += 4;
     outptr += 4;
   }
 
   // inverse transform column vectors
   for (i = 0; i < 4; ++i) {
     for (j = 0; j < 4; ++j)
       temp_in[j] = out[j * 4 + i];
     IHT_4[tx_type].cols(temp_in, temp_out);
     for (j = 0; j < 4; ++j)
-      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
-                                  + dest[j * dest_stride + i]);
+      dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
+                                  + dest[j * stride + i]);
   }
 }
-static void iadst8_1d(int16_t *input, int16_t *output) {
+static void iadst8_1d(const int16_t *input, int16_t *output) {
   int s0, s1, s2, s3, s4, s5, s6, s7;
 
   int x0 = input[7];
   int x1 = input[0];
   int x2 = input[5];
   int x3 = input[2];
   int x4 = input[3];
   int x5 = input[4];
   int x6 = input[1];
   int x7 = input[6];
@@ skipping 63 matching lines @@
   output[7] = -x1;
 }
 
 static const transform_2d IHT_8[] = {
   { idct8_1d,  idct8_1d  },  // DCT_DCT  = 0
   { iadst8_1d, idct8_1d  },  // ADST_DCT = 1
   { idct8_1d,  iadst8_1d },  // DCT_ADST = 2
   { iadst8_1d, iadst8_1d }   // ADST_ADST = 3
 };
 
-void vp9_short_iht8x8_add_c(int16_t *input, uint8_t *dest, int dest_stride,
-                            int tx_type) {
+void vp9_iht8x8_64_add_c(const int16_t *input, uint8_t *dest, int stride,
+                         int tx_type) {
   int i, j;
   int16_t out[8 * 8];
   int16_t *outptr = out;
   int16_t temp_in[8], temp_out[8];
   const transform_2d ht = IHT_8[tx_type];
 
   // inverse transform row vectors
   for (i = 0; i < 8; ++i) {
     ht.rows(input, outptr);
     input += 8;
     outptr += 8;
   }
 
   // inverse transform column vectors
   for (i = 0; i < 8; ++i) {
     for (j = 0; j < 8; ++j)
       temp_in[j] = out[j * 8 + i];
     ht.cols(temp_in, temp_out);
     for (j = 0; j < 8; ++j)
-      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
-                                  + dest[j * dest_stride + i]);  }
+      dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
+                                  + dest[j * stride + i]);
+  }
 }
 
-void vp9_short_idct10_8x8_add_c(int16_t *input, uint8_t *dest,
-                                int dest_stride) {
+void vp9_idct8x8_10_add_c(const int16_t *input, uint8_t *dest, int stride) {
   int16_t out[8 * 8] = { 0 };
   int16_t *outptr = out;
   int i, j;
   int16_t temp_in[8], temp_out[8];
 
   // First transform rows
   // only first 4 row has non-zero coefs
   for (i = 0; i < 4; ++i) {
     idct8_1d(input, outptr);
     input += 8;
     outptr += 8;
   }
 
   // Then transform columns
   for (i = 0; i < 8; ++i) {
     for (j = 0; j < 8; ++j)
       temp_in[j] = out[j * 8 + i];
     idct8_1d(temp_in, temp_out);
     for (j = 0; j < 8; ++j)
-      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
-                                  + dest[j * dest_stride + i]);
+      dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5)
+                                  + dest[j * stride + i]);
   }
 }
 
-static void idct16_1d(int16_t *input, int16_t *output) {
+static void idct16_1d(const int16_t *input, int16_t *output) {
   int16_t step1[16], step2[16];
   int temp1, temp2;
 
   // stage 1
   step1[0] = input[0/2];
   step1[1] = input[16/2];
   step1[2] = input[8/2];
   step1[3] = input[24/2];
   step1[4] = input[4/2];
   step1[5] = input[20/2];
@@ skipping 144 matching lines @@
   output[8] = step2[7] - step2[8];
   output[9] = step2[6] - step2[9];
   output[10] = step2[5] - step2[10];
   output[11] = step2[4] - step2[11];
   output[12] = step2[3] - step2[12];
   output[13] = step2[2] - step2[13];
   output[14] = step2[1] - step2[14];
   output[15] = step2[0] - step2[15];
 }
 
-void vp9_short_idct16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct16x16_256_add_c(const int16_t *input, uint8_t *dest, int stride) {
   int16_t out[16 * 16];
   int16_t *outptr = out;
   int i, j;
   int16_t temp_in[16], temp_out[16];
 
   // First transform rows
   for (i = 0; i < 16; ++i) {
     idct16_1d(input, outptr);
     input += 16;
     outptr += 16;
   }
 
   // Then transform columns
   for (i = 0; i < 16; ++i) {
     for (j = 0; j < 16; ++j)
       temp_in[j] = out[j * 16 + i];
     idct16_1d(temp_in, temp_out);
     for (j = 0; j < 16; ++j)
-      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
-                                  + dest[j * dest_stride + i]);
+      dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+                                  + dest[j * stride + i]);
   }
 }
 
-void iadst16_1d(int16_t *input, int16_t *output) {
+static void iadst16_1d(const int16_t *input, int16_t *output) {
   int s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12, s13, s14, s15;
 
   int x0 = input[15];
   int x1 = input[0];
   int x2 = input[13];
   int x3 = input[2];
   int x4 = input[11];
   int x5 = input[4];
   int x6 = input[9];
   int x7 = input[6];
@@ skipping 157 matching lines @@
   output[15] = -x1;
 }
 
 static const transform_2d IHT_16[] = {
   { idct16_1d,  idct16_1d  },  // DCT_DCT  = 0
   { iadst16_1d, idct16_1d  },  // ADST_DCT = 1
   { idct16_1d,  iadst16_1d },  // DCT_ADST = 2
   { iadst16_1d, iadst16_1d }   // ADST_ADST = 3
 };
 
-void vp9_short_iht16x16_add_c(int16_t *input, uint8_t *dest, int dest_stride,
-                              int tx_type) {
+void vp9_iht16x16_256_add_c(const int16_t *input, uint8_t *dest, int stride,
+                            int tx_type) {
   int i, j;
   int16_t out[16 * 16];
   int16_t *outptr = out;
   int16_t temp_in[16], temp_out[16];
   const transform_2d ht = IHT_16[tx_type];
 
   // Rows
   for (i = 0; i < 16; ++i) {
     ht.rows(input, outptr);
     input += 16;
     outptr += 16;
   }
 
   // Columns
   for (i = 0; i < 16; ++i) {
     for (j = 0; j < 16; ++j)
       temp_in[j] = out[j * 16 + i];
     ht.cols(temp_in, temp_out);
     for (j = 0; j < 16; ++j)
-      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
-                                  + dest[j * dest_stride + i]);  }
+      dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+                                  + dest[j * stride + i]);  }
 }
 
-void vp9_short_idct10_16x16_add_c(int16_t *input, uint8_t *dest,
-                                  int dest_stride) {
+void vp9_idct16x16_10_add_c(const int16_t *input, uint8_t *dest, int stride) {
   int16_t out[16 * 16] = { 0 };
   int16_t *outptr = out;
   int i, j;
   int16_t temp_in[16], temp_out[16];
 
   // First transform rows. Since all non-zero dct coefficients are in
   // upper-left 4x4 area, we only need to calculate first 4 rows here.
   for (i = 0; i < 4; ++i) {
     idct16_1d(input, outptr);
     input += 16;
     outptr += 16;
   }
 
   // Then transform columns
   for (i = 0; i < 16; ++i) {
     for (j = 0; j < 16; ++j)
       temp_in[j] = out[j*16 + i];
     idct16_1d(temp_in, temp_out);
     for (j = 0; j < 16; ++j)
-      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
-                                  + dest[j * dest_stride + i]);
+      dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+                                  + dest[j * stride + i]);
   }
 }
 
-void vp9_short_idct16x16_1_add_c(int16_t *input, uint8_t *dest,
-                                 int dest_stride) {
+void vp9_idct16x16_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
   int i, j;
   int a1;
   int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
   out = dct_const_round_shift(out * cospi_16_64);
   a1 = ROUND_POWER_OF_TWO(out, 6);
   for (j = 0; j < 16; ++j) {
     for (i = 0; i < 16; ++i)
       dest[i] = clip_pixel(dest[i] + a1);
-    dest += dest_stride;
+    dest += stride;
   }
 }
 
-static void idct32_1d(int16_t *input, int16_t *output) {
+static void idct32_1d(const int16_t *input, int16_t *output) {
   int16_t step1[32], step2[32];
   int temp1, temp2;
 
   // stage 1
   step1[0] = input[0];
   step1[1] = input[16];
   step1[2] = input[8];
   step1[3] = input[24];
   step1[4] = input[4];
   step1[5] = input[20];
@@ skipping 346 matching lines @@
   output[24] = step1[7] - step1[24];
   output[25] = step1[6] - step1[25];
   output[26] = step1[5] - step1[26];
   output[27] = step1[4] - step1[27];
   output[28] = step1[3] - step1[28];
   output[29] = step1[2] - step1[29];
   output[30] = step1[1] - step1[30];
   output[31] = step1[0] - step1[31];
 }
 
-void vp9_short_idct32x32_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+void vp9_idct32x32_1024_add_c(const int16_t *input, uint8_t *dest, int stride) {
   int16_t out[32 * 32];
   int16_t *outptr = out;
   int i, j;
   int16_t temp_in[32], temp_out[32];
 
   // Rows
   for (i = 0; i < 32; ++i) {
+    int16_t zero_coeff[16];
+    for (j = 0; j < 16; ++j)
+      zero_coeff[j] = input[2 * j] | input[2 * j + 1];
+    for (j = 0; j < 8; ++j)
+      zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+    for (j = 0; j < 4; ++j)
+      zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+    for (j = 0; j < 2; ++j)
+      zero_coeff[j] = zero_coeff[2 * j] | zero_coeff[2 * j + 1];
+
+    if (zero_coeff[0] | zero_coeff[1])
+      idct32_1d(input, outptr);
+    else
+      vpx_memset(outptr, 0, sizeof(int16_t) * 32);
+    input += 32;
+    outptr += 32;
+  }
+
+  // Columns
+  for (i = 0; i < 32; ++i) {
+    for (j = 0; j < 32; ++j)
+      temp_in[j] = out[j * 32 + i];
+    idct32_1d(temp_in, temp_out);
+    for (j = 0; j < 32; ++j)
+      dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+                                  + dest[j * stride + i]);
+  }
+}
+
+void vp9_idct32x32_34_add_c(const int16_t *input, uint8_t *dest, int stride) {
+  int16_t out[32 * 32] = {0};
+  int16_t *outptr = out;
+  int i, j;
+  int16_t temp_in[32], temp_out[32];
+
+  // Rows
+  // only upper-left 8x8 has non-zero coeff
+  for (i = 0; i < 8; ++i) {
     idct32_1d(input, outptr);
     input += 32;
     outptr += 32;
   }
 
   // Columns
   for (i = 0; i < 32; ++i) {
     for (j = 0; j < 32; ++j)
       temp_in[j] = out[j * 32 + i];
     idct32_1d(temp_in, temp_out);
     for (j = 0; j < 32; ++j)
-      dest[j * dest_stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
-                                  + dest[j * dest_stride + i]);
+      dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+                                  + dest[j * stride + i]);
   }
 }
 
-void vp9_short_idct1_32x32_c(int16_t *input, int16_t *output) {
+void vp9_idct32x32_1_add_c(const int16_t *input, uint8_t *dest, int stride) {
+  int i, j;
+  int a1;
+
   int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
   out = dct_const_round_shift(out * cospi_16_64);
-  output[0] = ROUND_POWER_OF_TWO(out, 6);
+  a1 = ROUND_POWER_OF_TWO(out, 6);
+
+  for (j = 0; j < 32; ++j) {
+    for (i = 0; i < 32; ++i)
+      dest[i] = clip_pixel(dest[i] + a1);
+    dest += stride;
+  }
 }
+
+// idct
+void vp9_idct4x4_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
+  if (eob > 1)
+    vp9_idct4x4_16_add(input, dest, stride);
+  else
+    vp9_idct4x4_1_add(input, dest, stride);
+}
+
+
+void vp9_iwht4x4_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
+  if (eob > 1)
+    vp9_iwht4x4_16_add(input, dest, stride);
+  else
+    vp9_iwht4x4_1_add(input, dest, stride);
+}
+
+void vp9_idct8x8_add(const int16_t *input, uint8_t *dest, int stride, int eob) {
+  // If dc is 1, then input[0] is the reconstructed value, do not need
+  // dequantization. Also, when dc is 1, dc is counted in eobs, namely eobs >=1.
+
+  // The calculation can be simplified if there are not many non-zero dct
+  // coefficients. Use eobs to decide what to do.
+  // TODO(yunqingwang): "eobs = 1" case is also handled in vp9_short_idct8x8_c.
+  // Combine that with code here.
+  if (eob) {
+    if (eob == 1)
+      // DC only DCT coefficient
+      vp9_idct8x8_1_add(input, dest, stride);
+    else if (eob <= 10)
+      vp9_idct8x8_10_add(input, dest, stride);
+    else
+      vp9_idct8x8_64_add(input, dest, stride);
+  }
+}
+
+void vp9_idct16x16_add(const int16_t *input, uint8_t *dest, int stride,
+                       int eob) {
+  /* The calculation can be simplified if there are not many non-zero dct
+   * coefficients. Use eobs to separate different cases. */
+  if (eob) {
+    if (eob == 1)
+      /* DC only DCT coefficient. */
+      vp9_idct16x16_1_add(input, dest, stride);
+    else if (eob <= 10)
+      vp9_idct16x16_10_add(input, dest, stride);
+    else
+      vp9_idct16x16_256_add(input, dest, stride);
+  }
+}
+
+void vp9_idct32x32_add(const int16_t *input, uint8_t *dest, int stride,
+                       int eob) {
+  if (eob) {
+    if (eob == 1)
+      vp9_idct32x32_1_add(input, dest, stride);
+    else if (eob <= 34)
+      // non-zero coeff only in upper-left 8x8
+      vp9_idct32x32_34_add(input, dest, stride);
+    else
+      vp9_idct32x32_1024_add(input, dest, stride);
+  }
+}
+
+// iht
+void vp9_iht4x4_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
+                    int stride, int eob) {
+  if (tx_type == DCT_DCT)
+    vp9_idct4x4_add(input, dest, stride, eob);
+  else
+    vp9_iht4x4_16_add(input, dest, stride, tx_type);
+}
+
+void vp9_iht8x8_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
+                    int stride, int eob) {
+  if (tx_type == DCT_DCT) {
+    vp9_idct8x8_add(input, dest, stride, eob);
+  } else {
+    if (eob > 0) {
+      vp9_iht8x8_64_add(input, dest, stride, tx_type);
+    }
+  }
+}
+
+void vp9_iht16x16_add(TX_TYPE tx_type, const int16_t *input, uint8_t *dest,
+                      int stride, int eob) {
+  if (tx_type == DCT_DCT) {
+    vp9_idct16x16_add(input, dest, stride, eob);
+  } else {
+    if (eob > 0) {
+      vp9_iht16x16_256_add(input, dest, stride, tx_type);
+    }
+  }
+}