libvpx: Pull from upstream

source/libvpx/vp9/encoder/vp9_dct.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/common/vp9_systemdependent.h"
+#include "./vp9_rtcd.h"
 
 #include "vp9/common/vp9_blockd.h"
 #include "vp9/common/vp9_idct.h"
+#include "vp9/common/vp9_systemdependent.h"
 
-static void fdct4_1d(int16_t *input, int16_t *output) {
+#include "vp9/encoder/vp9_dct.h"
+
+static void fdct4(const int16_t *input, int16_t *output) {
   int16_t step[4];
   int temp1, temp2;
 
   step[0] = input[0] + input[3];
   step[1] = input[1] + input[2];
   step[2] = input[1] - input[2];
   step[3] = input[0] - input[3];
 
   temp1 = (step[0] + step[1]) * cospi_16_64;
   temp2 = (step[0] - step[1]) * cospi_16_64;
   output[0] = dct_const_round_shift(temp1);
   output[2] = dct_const_round_shift(temp2);
   temp1 = step[2] * cospi_24_64 + step[3] * cospi_8_64;
   temp2 = -step[2] * cospi_8_64 + step[3] * cospi_24_64;
   output[1] = dct_const_round_shift(temp1);
   output[3] = dct_const_round_shift(temp2);
 }
 
-void vp9_short_fdct4x4_c(int16_t *input, int16_t *output, int pitch) {
+void vp9_fdct4x4_c(const int16_t *input, int16_t *output, int stride) {
   // The 2D transform is done with two passes which are actually pretty
   // similar. In the first one, we transform the columns and transpose
   // the results. In the second one, we transform the rows. To achieve that,
   // as the first pass results are transposed, we tranpose the columns (that
   // is the transposed rows) and transpose the results (so that it goes back
   // in normal/row positions).
-  const int stride = pitch >> 1;
   int pass;
   // We need an intermediate buffer between passes.
   int16_t intermediate[4 * 4];
-  int16_t *in = input;
+  const int16_t *in = input;
   int16_t *out = intermediate;
   // Do the two transform/transpose passes
   for (pass = 0; pass < 2; ++pass) {
     /*canbe16*/ int input[4];
     /*canbe16*/ int step[4];
     /*needs32*/ int temp1, temp2;
     int i;
     for (i = 0; i < 4; ++i) {
       // Load inputs.
       if (0 == pass) {
-        input[0] = in[0 * stride] << 4;
-        input[1] = in[1 * stride] << 4;
-        input[2] = in[2 * stride] << 4;
-        input[3] = in[3 * stride] << 4;
+        input[0] = in[0 * stride] * 16;
+        input[1] = in[1 * stride] * 16;
+        input[2] = in[2 * stride] * 16;
+        input[3] = in[3 * stride] * 16;
         if (i == 0 && input[0]) {
           input[0] += 1;
         }
       } else {
         input[0] = in[0 * 4];
         input[1] = in[1 * 4];
         input[2] = in[2 * 4];
         input[3] = in[3 * 4];
       }
       // Transform.
@@ skipping 20 matching lines @@
 
   {
     int i, j;
     for (i = 0; i < 4; ++i) {
       for (j = 0; j < 4; ++j)
         output[j + i * 4] = (output[j + i * 4] + 1) >> 2;
     }
   }
 }
 
-static void fadst4_1d(int16_t *input, int16_t *output) {
+static void fadst4(const int16_t *input, int16_t *output) {
   int x0, x1, x2, x3;
   int s0, s1, s2, s3, s4, s5, s6, s7;
 
   x0 = input[0];
   x1 = input[1];
   x2 = input[2];
   x3 = input[3];
 
   if (!(x0 | x1 | x2 | x3)) {
     output[0] = output[1] = output[2] = output[3] = 0;
@@ skipping 20 matching lines @@
   s3 = x2 - x0 + x3;
 
   // 1-D transform scaling factor is sqrt(2).
   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);
 }
 
 static const transform_2d FHT_4[] = {
-  { fdct4_1d,  fdct4_1d  },  // DCT_DCT  = 0
-  { fadst4_1d, fdct4_1d  },  // ADST_DCT = 1
-  { fdct4_1d,  fadst4_1d },  // DCT_ADST = 2
-  { fadst4_1d, fadst4_1d }   // ADST_ADST = 3
+  { fdct4,  fdct4  },  // DCT_DCT  = 0
+  { fadst4, fdct4  },  // ADST_DCT = 1
+  { fdct4,  fadst4 },  // DCT_ADST = 2
+  { fadst4, fadst4 }   // ADST_ADST = 3
 };
 
-void vp9_short_fht4x4_c(int16_t *input, int16_t *output,
-                        int pitch, TX_TYPE tx_type) {
+void vp9_short_fht4x4_c(const int16_t *input, int16_t *output,
+                        int stride, int tx_type) {
   int16_t out[4 * 4];
   int16_t *outptr = &out[0];
   int i, j;
   int16_t temp_in[4], temp_out[4];
   const transform_2d ht = FHT_4[tx_type];
 
   // Columns
   for (i = 0; i < 4; ++i) {
     for (j = 0; j < 4; ++j)
-      temp_in[j] = input[j * pitch + i] << 4;
+      temp_in[j] = input[j * stride + i] * 16;
     if (i == 0 && temp_in[0])
       temp_in[0] += 1;
     ht.cols(temp_in, temp_out);
     for (j = 0; j < 4; ++j)
       outptr[j * 4 + i] = temp_out[j];
   }
 
   // Rows
   for (i = 0; i < 4; ++i) {
     for (j = 0; j < 4; ++j)
       temp_in[j] = out[j + i * 4];
     ht.rows(temp_in, temp_out);
     for (j = 0; j < 4; ++j)
       output[j + i * 4] = (temp_out[j] + 1) >> 2;
   }
 }
 
-void vp9_short_fdct8x4_c(int16_t *input, int16_t *output, int pitch) {
-    vp9_short_fdct4x4_c(input, output, pitch);
-    vp9_short_fdct4x4_c(input + 4, output + 16, pitch);
-}
-
-static void fdct8_1d(int16_t *input, int16_t *output) {
+static void fdct8(const int16_t *input, int16_t *output) {
   /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
   /*needs32*/ int t0, t1, t2, t3;
   /*canbe16*/ int x0, x1, x2, x3;
 
   // stage 1
   s0 = input[0] + input[7];
   s1 = input[1] + input[6];
   s2 = input[2] + input[5];
   s3 = input[3] + input[4];
   s4 = input[3] - input[4];
   s5 = input[2] - input[5];
   s6 = input[1] - input[6];
   s7 = input[0] - input[7];
 
-  // fdct4_1d(step, step);
+  // fdct4(step, step);
   x0 = s0 + s3;
   x1 = s1 + s2;
   x2 = s1 - s2;
   x3 = s0 - s3;
   t0 = (x0 + x1) * cospi_16_64;
   t1 = (x0 - x1) * cospi_16_64;
   t2 =  x2 * cospi_24_64 + x3 *  cospi_8_64;
   t3 = -x2 * cospi_8_64  + x3 * cospi_24_64;
   output[0] = dct_const_round_shift(t0);
   output[2] = dct_const_round_shift(t2);
@@ skipping 16 matching lines @@
   t0 = x0 * cospi_28_64 + x3 *   cospi_4_64;
   t1 = x1 * cospi_12_64 + x2 *  cospi_20_64;
   t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
   t3 = x3 * cospi_28_64 + x0 *  -cospi_4_64;
   output[1] = dct_const_round_shift(t0);
   output[3] = dct_const_round_shift(t2);
   output[5] = dct_const_round_shift(t1);
   output[7] = dct_const_round_shift(t3);
 }
 
-void vp9_short_fdct8x8_c(int16_t *input, int16_t *final_output, int pitch) {
-  const int stride = pitch >> 1;
+void vp9_fdct8x8_c(const int16_t *input, int16_t *final_output, int stride) {
   int i, j;
   int16_t intermediate[64];
 
   // Transform columns
   {
     int16_t *output = intermediate;
     /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
     /*needs32*/ int t0, t1, t2, t3;
     /*canbe16*/ int x0, x1, x2, x3;
 
     int i;
     for (i = 0; i < 8; i++) {
       // stage 1
-      s0 = (input[0 * stride] + input[7 * stride]) << 2;
-      s1 = (input[1 * stride] + input[6 * stride]) << 2;
-      s2 = (input[2 * stride] + input[5 * stride]) << 2;
-      s3 = (input[3 * stride] + input[4 * stride]) << 2;
-      s4 = (input[3 * stride] - input[4 * stride]) << 2;
-      s5 = (input[2 * stride] - input[5 * stride]) << 2;
-      s6 = (input[1 * stride] - input[6 * stride]) << 2;
-      s7 = (input[0 * stride] - input[7 * stride]) << 2;
+      s0 = (input[0 * stride] + input[7 * stride]) * 4;
+      s1 = (input[1 * stride] + input[6 * stride]) * 4;
+      s2 = (input[2 * stride] + input[5 * stride]) * 4;
+      s3 = (input[3 * stride] + input[4 * stride]) * 4;
+      s4 = (input[3 * stride] - input[4 * stride]) * 4;
+      s5 = (input[2 * stride] - input[5 * stride]) * 4;
+      s6 = (input[1 * stride] - input[6 * stride]) * 4;
+      s7 = (input[0 * stride] - input[7 * stride]) * 4;
 
-      // fdct4_1d(step, step);
+      // fdct4(step, step);
       x0 = s0 + s3;
       x1 = s1 + s2;
       x2 = s1 - s2;
       x3 = s0 - s3;
       t0 = (x0 + x1) * cospi_16_64;
       t1 = (x0 - x1) * cospi_16_64;
       t2 =  x2 * cospi_24_64 + x3 *  cospi_8_64;
       t3 = -x2 * cospi_8_64  + x3 * cospi_24_64;
       output[0 * 8] = dct_const_round_shift(t0);
       output[2 * 8] = dct_const_round_shift(t2);
@@ skipping 21 matching lines @@
       output[3 * 8] = dct_const_round_shift(t2);
       output[5 * 8] = dct_const_round_shift(t1);
       output[7 * 8] = dct_const_round_shift(t3);
       input++;
       output++;
     }
   }
 
   // Rows
   for (i = 0; i < 8; ++i) {
-    fdct8_1d(&intermediate[i * 8], &final_output[i * 8]);
+    fdct8(&intermediate[i * 8], &final_output[i * 8]);
     for (j = 0; j < 8; ++j)
       final_output[j + i * 8] /= 2;
   }
 }
 
-void vp9_short_fdct16x16_c(int16_t *input, int16_t *output, int pitch) {
+void vp9_fdct16x16_c(const int16_t *input, int16_t *output, int stride) {
   // The 2D transform is done with two passes which are actually pretty
   // similar. In the first one, we transform the columns and transpose
   // the results. In the second one, we transform the rows. To achieve that,
   // as the first pass results are transposed, we tranpose the columns (that
   // is the transposed rows) and transpose the results (so that it goes back
   // in normal/row positions).
-  const int stride = pitch >> 1;
   int pass;
   // We need an intermediate buffer between passes.
   int16_t intermediate[256];
-  int16_t *in = input;
+  const int16_t *in = input;
   int16_t *out = intermediate;
   // Do the two transform/transpose passes
   for (pass = 0; pass < 2; ++pass) {
     /*canbe16*/ int step1[8];
     /*canbe16*/ int step2[8];
     /*canbe16*/ int step3[8];
     /*canbe16*/ int input[8];
     /*needs32*/ int temp1, temp2;
     int i;
     for (i = 0; i < 16; i++) {
       if (0 == pass) {
         // Calculate input for the first 8 results.
-        input[0] = (in[0 * stride] + in[15 * stride]) << 2;
-        input[1] = (in[1 * stride] + in[14 * stride]) << 2;
-        input[2] = (in[2 * stride] + in[13 * stride]) << 2;
-        input[3] = (in[3 * stride] + in[12 * stride]) << 2;
-        input[4] = (in[4 * stride] + in[11 * stride]) << 2;
-        input[5] = (in[5 * stride] + in[10 * stride]) << 2;
-        input[6] = (in[6 * stride] + in[ 9 * stride]) << 2;
-        input[7] = (in[7 * stride] + in[ 8 * stride]) << 2;
+        input[0] = (in[0 * stride] + in[15 * stride]) * 4;
+        input[1] = (in[1 * stride] + in[14 * stride]) * 4;
+        input[2] = (in[2 * stride] + in[13 * stride]) * 4;
+        input[3] = (in[3 * stride] + in[12 * stride]) * 4;
+        input[4] = (in[4 * stride] + in[11 * stride]) * 4;
+        input[5] = (in[5 * stride] + in[10 * stride]) * 4;
+        input[6] = (in[6 * stride] + in[ 9 * stride]) * 4;
+        input[7] = (in[7 * stride] + in[ 8 * stride]) * 4;
         // Calculate input for the next 8 results.
-        step1[0] = (in[7 * stride] - in[ 8 * stride]) << 2;
-        step1[1] = (in[6 * stride] - in[ 9 * stride]) << 2;
-        step1[2] = (in[5 * stride] - in[10 * stride]) << 2;
-        step1[3] = (in[4 * stride] - in[11 * stride]) << 2;
-        step1[4] = (in[3 * stride] - in[12 * stride]) << 2;
-        step1[5] = (in[2 * stride] - in[13 * stride]) << 2;
-        step1[6] = (in[1 * stride] - in[14 * stride]) << 2;
-        step1[7] = (in[0 * stride] - in[15 * stride]) << 2;
+        step1[0] = (in[7 * stride] - in[ 8 * stride]) * 4;
+        step1[1] = (in[6 * stride] - in[ 9 * stride]) * 4;
+        step1[2] = (in[5 * stride] - in[10 * stride]) * 4;
+        step1[3] = (in[4 * stride] - in[11 * stride]) * 4;
+        step1[4] = (in[3 * stride] - in[12 * stride]) * 4;
+        step1[5] = (in[2 * stride] - in[13 * stride]) * 4;
+        step1[6] = (in[1 * stride] - in[14 * stride]) * 4;
+        step1[7] = (in[0 * stride] - in[15 * stride]) * 4;
       } else {
         // Calculate input for the first 8 results.
         input[0] = ((in[0 * 16] + 1) >> 2) + ((in[15 * 16] + 1) >> 2);
         input[1] = ((in[1 * 16] + 1) >> 2) + ((in[14 * 16] + 1) >> 2);
         input[2] = ((in[2 * 16] + 1) >> 2) + ((in[13 * 16] + 1) >> 2);
         input[3] = ((in[3 * 16] + 1) >> 2) + ((in[12 * 16] + 1) >> 2);
         input[4] = ((in[4 * 16] + 1) >> 2) + ((in[11 * 16] + 1) >> 2);
         input[5] = ((in[5 * 16] + 1) >> 2) + ((in[10 * 16] + 1) >> 2);
         input[6] = ((in[6 * 16] + 1) >> 2) + ((in[ 9 * 16] + 1) >> 2);
         input[7] = ((in[7 * 16] + 1) >> 2) + ((in[ 8 * 16] + 1) >> 2);
         // Calculate input for the next 8 results.
         step1[0] = ((in[7 * 16] + 1) >> 2) - ((in[ 8 * 16] + 1) >> 2);
         step1[1] = ((in[6 * 16] + 1) >> 2) - ((in[ 9 * 16] + 1) >> 2);
         step1[2] = ((in[5 * 16] + 1) >> 2) - ((in[10 * 16] + 1) >> 2);
         step1[3] = ((in[4 * 16] + 1) >> 2) - ((in[11 * 16] + 1) >> 2);
         step1[4] = ((in[3 * 16] + 1) >> 2) - ((in[12 * 16] + 1) >> 2);
         step1[5] = ((in[2 * 16] + 1) >> 2) - ((in[13 * 16] + 1) >> 2);
         step1[6] = ((in[1 * 16] + 1) >> 2) - ((in[14 * 16] + 1) >> 2);
         step1[7] = ((in[0 * 16] + 1) >> 2) - ((in[15 * 16] + 1) >> 2);
       }
-      // Work on the first eight values; fdct8_1d(input, even_results);
+      // Work on the first eight values; fdct8(input, even_results);
       {
         /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
         /*needs32*/ int t0, t1, t2, t3;
         /*canbe16*/ int x0, x1, x2, x3;
 
         // stage 1
         s0 = input[0] + input[7];
         s1 = input[1] + input[6];
         s2 = input[2] + input[5];
         s3 = input[3] + input[4];
         s4 = input[3] - input[4];
         s5 = input[2] - input[5];
         s6 = input[1] - input[6];
         s7 = input[0] - input[7];
 
-        // fdct4_1d(step, step);
+        // fdct4(step, step);
         x0 = s0 + s3;
         x1 = s1 + s2;
         x2 = s1 - s2;
         x3 = s0 - s3;
         t0 = (x0 + x1) * cospi_16_64;
         t1 = (x0 - x1) * cospi_16_64;
         t2 = x3 * cospi_8_64  + x2 * cospi_24_64;
         t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
         out[0] = dct_const_round_shift(t0);
         out[4] = dct_const_round_shift(t2);
@@ skipping 81 matching lines @@
       // Do next column (which is a transposed row in second/horizontal pass)
       in++;
       out += 16;
     }
     // Setup in/out for next pass.
     in = intermediate;
     out = output;
   }
 }
 
-static void fadst8_1d(int16_t *input, int16_t *output) {
+static void fadst8(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 51 matching lines @@
   output[1] = - x4;
   output[2] =   x6;
   output[3] = - x2;
   output[4] =   x3;
   output[5] = - x7;
   output[6] =   x5;
   output[7] = - x1;
 }
 
 static const transform_2d FHT_8[] = {
-  { fdct8_1d,  fdct8_1d  },  // DCT_DCT  = 0
-  { fadst8_1d, fdct8_1d  },  // ADST_DCT = 1
-  { fdct8_1d,  fadst8_1d },  // DCT_ADST = 2
-  { fadst8_1d, fadst8_1d }   // ADST_ADST = 3
+  { fdct8,  fdct8  },  // DCT_DCT  = 0
+  { fadst8, fdct8  },  // ADST_DCT = 1
+  { fdct8,  fadst8 },  // DCT_ADST = 2
+  { fadst8, fadst8 }   // ADST_ADST = 3
 };
 
-void vp9_short_fht8x8_c(int16_t *input, int16_t *output,
-                        int pitch, TX_TYPE tx_type) {
+void vp9_short_fht8x8_c(const int16_t *input, int16_t *output,
+                        int stride, int tx_type) {
   int16_t out[64];
   int16_t *outptr = &out[0];
   int i, j;
   int16_t temp_in[8], temp_out[8];
   const transform_2d ht = FHT_8[tx_type];
 
   // Columns
   for (i = 0; i < 8; ++i) {
     for (j = 0; j < 8; ++j)
-      temp_in[j] = input[j * pitch + i] << 2;
+      temp_in[j] = input[j * stride + i] * 4;
     ht.cols(temp_in, temp_out);
     for (j = 0; j < 8; ++j)
       outptr[j * 8 + i] = temp_out[j];
   }
 
   // Rows
   for (i = 0; i < 8; ++i) {
     for (j = 0; j < 8; ++j)
       temp_in[j] = out[j + i * 8];
     ht.rows(temp_in, temp_out);
     for (j = 0; j < 8; ++j)
       output[j + i * 8] = (temp_out[j] + (temp_out[j] < 0)) >> 1;
   }
 }
 
 /* 4-point reversible, orthonormal Walsh-Hadamard in 3.5 adds, 0.5 shifts per
    pixel. */
-void vp9_short_walsh4x4_c(short *input, short *output, int pitch) {
+void vp9_fwht4x4_c(const int16_t *input, int16_t *output, int stride) {
   int i;
   int a1, b1, c1, d1, e1;
-  short *ip = input;
-  short *op = output;
-  int pitch_short = pitch >> 1;
+  const int16_t *ip = input;
+  int16_t *op = output;
 
   for (i = 0; i < 4; i++) {
-    a1 = ip[0 * pitch_short];
-    b1 = ip[1 * pitch_short];
-    c1 = ip[2 * pitch_short];
-    d1 = ip[3 * pitch_short];
+    a1 = ip[0 * stride];
+    b1 = ip[1 * stride];
+    c1 = ip[2 * stride];
+    d1 = ip[3 * stride];
 
     a1 += b1;
     d1 = d1 - c1;
     e1 = (a1 - d1) >> 1;
     b1 = e1 - b1;
     c1 = e1 - c1;
     a1 -= c1;
     d1 += b1;
     op[0] = a1;
     op[4] = c1;
@@ skipping 12 matching lines @@
     c1 = ip[2];
     d1 = ip[3];
 
     a1 += b1;
     d1 -= c1;
     e1 = (a1 - d1) >> 1;
     b1 = e1 - b1;
     c1 = e1 - c1;
     a1 -= c1;
     d1 += b1;
-    op[0] = a1 << WHT_UPSCALE_FACTOR;
-    op[1] = c1 << WHT_UPSCALE_FACTOR;
-    op[2] = d1 << WHT_UPSCALE_FACTOR;
-    op[3] = b1 << WHT_UPSCALE_FACTOR;
+    op[0] = a1 * UNIT_QUANT_FACTOR;
+    op[1] = c1 * UNIT_QUANT_FACTOR;
+    op[2] = d1 * UNIT_QUANT_FACTOR;
+    op[3] = b1 * UNIT_QUANT_FACTOR;
 
     ip += 4;
     op += 4;
   }
 }
 
-void vp9_short_walsh8x4_c(short *input, short *output, int pitch) {
-  vp9_short_walsh4x4_c(input,   output,    pitch);
-  vp9_short_walsh4x4_c(input + 4, output + 16, pitch);
-}
-
-
 // Rewrote to use same algorithm as others.
-static void fdct16_1d(int16_t in[16], int16_t out[16]) {
+static void fdct16(const int16_t in[16], int16_t out[16]) {
   /*canbe16*/ int step1[8];
   /*canbe16*/ int step2[8];
   /*canbe16*/ int step3[8];
   /*canbe16*/ int input[8];
   /*needs32*/ int temp1, temp2;
 
   // step 1
   input[0] = in[0] + in[15];
   input[1] = in[1] + in[14];
   input[2] = in[2] + in[13];
   input[3] = in[3] + in[12];
   input[4] = in[4] + in[11];
   input[5] = in[5] + in[10];
   input[6] = in[6] + in[ 9];
   input[7] = in[7] + in[ 8];
 
   step1[0] = in[7] - in[ 8];
   step1[1] = in[6] - in[ 9];
   step1[2] = in[5] - in[10];
   step1[3] = in[4] - in[11];
   step1[4] = in[3] - in[12];
   step1[5] = in[2] - in[13];
   step1[6] = in[1] - in[14];
   step1[7] = in[0] - in[15];
 
-  // fdct8_1d(step, step);
+  // fdct8(step, step);
   {
     /*canbe16*/ int s0, s1, s2, s3, s4, s5, s6, s7;
     /*needs32*/ int t0, t1, t2, t3;
     /*canbe16*/ int x0, x1, x2, x3;
 
     // stage 1
     s0 = input[0] + input[7];
     s1 = input[1] + input[6];
     s2 = input[2] + input[5];
     s3 = input[3] + input[4];
     s4 = input[3] - input[4];
     s5 = input[2] - input[5];
     s6 = input[1] - input[6];
     s7 = input[0] - input[7];
 
-    // fdct4_1d(step, step);
+    // fdct4(step, step);
     x0 = s0 + s3;
     x1 = s1 + s2;
     x2 = s1 - s2;
     x3 = s0 - s3;
     t0 = (x0 + x1) * cospi_16_64;
     t1 = (x0 - x1) * cospi_16_64;
     t2 = x3 * cospi_8_64  + x2 * cospi_24_64;
     t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
     out[0] = dct_const_round_shift(t0);
     out[4] = dct_const_round_shift(t2);
@@ skipping 78 matching lines @@
   temp2 = step1[2] * -cospi_10_64 + step1[5] * cospi_22_64;
   out[3] = dct_const_round_shift(temp1);
   out[11] = dct_const_round_shift(temp2);
 
   temp1 = step1[1] * -cospi_18_64 + step1[6] * cospi_14_64;
   temp2 = step1[0] *  -cospi_2_64 + step1[7] * cospi_30_64;
   out[7] = dct_const_round_shift(temp1);
   out[15] = dct_const_round_shift(temp2);
 }
 
-void fadst16_1d(int16_t *input, int16_t *output) {
+static void fadst16(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 142 matching lines @@
   output[9] =  x11;
   output[10] = x15;
   output[11] = x7;
   output[12] = x5;
   output[13] = - x13;
   output[14] = x9;
   output[15] = - x1;
 }
 
 static const transform_2d FHT_16[] = {
-  { fdct16_1d,  fdct16_1d  },  // DCT_DCT  = 0
-  { fadst16_1d, fdct16_1d  },  // ADST_DCT = 1
-  { fdct16_1d,  fadst16_1d },  // DCT_ADST = 2
-  { fadst16_1d, fadst16_1d }   // ADST_ADST = 3
+  { fdct16,  fdct16  },  // DCT_DCT  = 0
+  { fadst16, fdct16  },  // ADST_DCT = 1
+  { fdct16,  fadst16 },  // DCT_ADST = 2
+  { fadst16, fadst16 }   // ADST_ADST = 3
 };
 
-void vp9_short_fht16x16_c(int16_t *input, int16_t *output,
-                          int pitch, TX_TYPE tx_type) {
+void vp9_short_fht16x16_c(const int16_t *input, int16_t *output,
+                          int stride, int tx_type) {
   int16_t out[256];
   int16_t *outptr = &out[0];
   int i, j;
   int16_t temp_in[16], temp_out[16];
   const transform_2d ht = FHT_16[tx_type];
 
   // Columns
   for (i = 0; i < 16; ++i) {
     for (j = 0; j < 16; ++j)
-      temp_in[j] = input[j * pitch + i] << 2;
+      temp_in[j] = input[j * stride + i] * 4;
     ht.cols(temp_in, temp_out);
     for (j = 0; j < 16; ++j)
       outptr[j * 16 + i] = (temp_out[j] + 1 + (temp_out[j] < 0)) >> 2;
 //      outptr[j * 16 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
   }
 
   // Rows
   for (i = 0; i < 16; ++i) {
     for (j = 0; j < 16; ++j)
       temp_in[j] = out[j + i * 16];
     ht.rows(temp_in, temp_out);
     for (j = 0; j < 16; ++j)
       output[j + i * 16] = temp_out[j];
   }
 }
 
 static INLINE int dct_32_round(int input) {
   int rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
   assert(-131072 <= rv && rv <= 131071);
   return rv;
 }
 
 static INLINE int half_round_shift(int input) {
   int rv = (input + 1 + (input < 0)) >> 2;
   return rv;
 }
 
-static void dct32_1d(int *input, int *output, int round) {
+static void dct32_1d(const int *input, int *output, int round) {
   int step[32];
   // Stage 1
   step[0] = input[0] + input[(32 - 1)];
   step[1] = input[1] + input[(32 - 2)];
   step[2] = input[2] + input[(32 - 3)];
   step[3] = input[3] + input[(32 - 4)];
   step[4] = input[4] + input[(32 - 5)];
   step[5] = input[5] + input[(32 - 6)];
   step[6] = input[6] + input[(32 - 7)];
   step[7] = input[7] + input[(32 - 8)];
@@ skipping 302 matching lines @@
   output[3]  = dct_32_round(step[24] * cospi_3_64 + step[23] * -cospi_29_64);
   output[19] = dct_32_round(step[25] * cospi_19_64 + step[22] * -cospi_13_64);
   output[11] = dct_32_round(step[26] * cospi_11_64 + step[21] * -cospi_21_64);
   output[27] = dct_32_round(step[27] * cospi_27_64 + step[20] * -cospi_5_64);
   output[7]  = dct_32_round(step[28] * cospi_7_64 + step[19] * -cospi_25_64);
   output[23] = dct_32_round(step[29] * cospi_23_64 + step[18] * -cospi_9_64);
   output[15] = dct_32_round(step[30] * cospi_15_64 + step[17] * -cospi_17_64);
   output[31] = dct_32_round(step[31] * cospi_31_64 + step[16] * -cospi_1_64);
 }
 
-void vp9_short_fdct32x32_c(int16_t *input, int16_t *out, int pitch) {
-  int shortpitch = pitch >> 1;
+void vp9_fdct32x32_c(const int16_t *input, int16_t *out, int stride) {
   int i, j;
   int output[32 * 32];
 
   // Columns
   for (i = 0; i < 32; ++i) {
     int temp_in[32], temp_out[32];
     for (j = 0; j < 32; ++j)
-      temp_in[j] = input[j * shortpitch + i] << 2;
+      temp_in[j] = input[j * stride + i] * 4;
     dct32_1d(temp_in, temp_out, 0);
     for (j = 0; j < 32; ++j)
       output[j * 32 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
   }
 
   // Rows
   for (i = 0; i < 32; ++i) {
     int temp_in[32], temp_out[32];
     for (j = 0; j < 32; ++j)
       temp_in[j] = output[j + i * 32];
     dct32_1d(temp_in, temp_out, 0);
     for (j = 0; j < 32; ++j)
       out[j + i * 32] = (temp_out[j] + 1 + (temp_out[j] < 0)) >> 2;
   }
 }
 
 // Note that although we use dct_32_round in dct32_1d computation flow,
 // this 2d fdct32x32 for rate-distortion optimization loop is operating
 // within 16 bits precision.
-void vp9_short_fdct32x32_rd_c(int16_t *input, int16_t *out, int pitch) {
-  int shortpitch = pitch >> 1;
+void vp9_fdct32x32_rd_c(const int16_t *input, int16_t *out, int stride) {
   int i, j;
   int output[32 * 32];
 
   // Columns
   for (i = 0; i < 32; ++i) {
     int temp_in[32], temp_out[32];
     for (j = 0; j < 32; ++j)
-      temp_in[j] = input[j * shortpitch + i] << 2;
+      temp_in[j] = input[j * stride + i] * 4;
     dct32_1d(temp_in, temp_out, 0);
     for (j = 0; j < 32; ++j)
       // TODO(cd): see quality impact of only doing
       //           output[j * 32 + i] = (temp_out[j] + 1) >> 2;
       //           PS: also change code in vp9/encoder/x86/vp9_dct_sse2.c
       output[j * 32 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
   }
 
   // Rows
   for (i = 0; i < 32; ++i) {
     int temp_in[32], temp_out[32];
     for (j = 0; j < 32; ++j)
       temp_in[j] = output[j + i * 32];
     dct32_1d(temp_in, temp_out, 1);
     for (j = 0; j < 32; ++j)
       out[j + i * 32] = temp_out[j];
   }
 }
+
+void vp9_fht4x4(TX_TYPE tx_type, const int16_t *input, int16_t *output,
+                int stride) {
+  if (tx_type == DCT_DCT)
+    vp9_fdct4x4(input, output, stride);
+  else
+    vp9_short_fht4x4(input, output, stride, tx_type);
+}
+
+void vp9_fht8x8(TX_TYPE tx_type, const int16_t *input, int16_t *output,
+                int stride) {
+  if (tx_type == DCT_DCT)
+    vp9_fdct8x8(input, output, stride);
+  else
+    vp9_short_fht8x8(input, output, stride, tx_type);
+}
+
+void vp9_fht16x16(TX_TYPE tx_type, const int16_t *input, int16_t *output,
+                  int stride) {
+  if (tx_type == DCT_DCT)
+    vp9_fdct16x16(input, output, stride);
+  else
+    vp9_short_fht16x16(input, output, stride, tx_type);
+}