shithub: libvpx

Download patch

ref: 9bf3f7516884b9baf22d21431a461752bc06c379
parent: 9770d564f4984e6a0d3cfdfb7e5b8bc83f52dccf
author: Dmitry Kovalev <dkovalev@google.com>
date: Mon Feb 25 13:19:55 EST 2013 

Changing pitch value meaning for fht and iht transforms.

Pitch now means the number of elements, not the number of bytes.

Change-Id: Idb9f2f012e39b09d596a3cc1802305a80b7c13af


--- a/vp9/common/vp9_idctllm.c
+++ b/vp9/common/vp9_idctllm.c
@@ -122,7 +122,7 @@
   }
 }
 
-void idct4_1d(int16_t *input, int16_t *output) {
+static void idct4_1d(int16_t *input, int16_t *output) {
   int16_t step[4];
   int temp1, temp2;
   // stage 1
@@ -200,7 +200,7 @@
   }
 }
 
-void idct8_1d(int16_t *input, int16_t *output) {
+static void idct8_1d(int16_t *input, int16_t *output) {
   int16_t step1[8], step2[8];
   int temp1, temp2;
   // stage 1
@@ -320,10 +320,9 @@
 
 void vp9_short_iht4x4_c(int16_t *input, int16_t *output,
                         int pitch, TX_TYPE tx_type) {
+  int i, j;
   int16_t out[4 * 4];
   int16_t *outptr = out;
-  const int half_pitch = pitch >> 1;
-  int i, j;
   int16_t temp_in[4], temp_out[4];
   const transform_2d ht = IHT_4[tx_type];
 
@@ -340,7 +339,7 @@
       temp_in[j] = out[j * 4 + i];
     ht.cols(temp_in, temp_out);
     for (j = 0; j < 4; ++j)
-      output[j * half_pitch + i] = ROUND_POWER_OF_TWO(temp_out[j], 4);
+      output[j * pitch + i] = ROUND_POWER_OF_TWO(temp_out[j], 4);
   }
 }
 
@@ -430,10 +429,9 @@
 
 void vp9_short_iht8x8_c(int16_t *input, int16_t *output,
                         int pitch, TX_TYPE tx_type) {
+  int i, j;
   int16_t out[8 * 8];
   int16_t *outptr = out;
-  const int half_pitch = pitch >> 1;
-  int i, j;
   int16_t temp_in[8], temp_out[8];
   const transform_2d ht = IHT_8[tx_type];
 
@@ -450,7 +448,7 @@
       temp_in[j] = out[j * 8 + i];
     ht.cols(temp_in, temp_out);
     for (j = 0; j < 8; ++j)
-      output[j * half_pitch + i] = ROUND_POWER_OF_TWO(temp_out[j], 5);
+      output[j * pitch + i] = ROUND_POWER_OF_TWO(temp_out[j], 5);
   }
 }
 
@@ -486,7 +484,7 @@
   output[0] = ROUND_POWER_OF_TWO(out, 5);
 }
 
-void idct16_1d(int16_t *input, int16_t *output) {
+static void idct16_1d(int16_t *input, int16_t *output) {
   int16_t step1[16], step2[16];
   int temp1, temp2;
 
@@ -853,11 +851,10 @@
 };
 
 void vp9_short_iht16x16_c(int16_t *input, int16_t *output,
-                          int pitch, TX_TYPE tx_type) {
+                          int input_pitch, TX_TYPE tx_type) {
+  int i, j;
   int16_t out[16 * 16];
   int16_t *outptr = out;
-  const int half_pitch = pitch >> 1;
-  int i, j;
   int16_t temp_in[16], temp_out[16];
   const transform_2d ht = IHT_16[tx_type];
 
@@ -864,7 +861,7 @@
   // Rows
   for (i = 0; i < 16; ++i) {
     ht.rows(input, outptr);
-    input += half_pitch;
+    input += input_pitch;
     outptr += 16;
   }
 
@@ -912,7 +909,7 @@
   output[0] = ROUND_POWER_OF_TWO(out, 6);
 }
 
-void idct32_1d(int16_t *input, int16_t *output) {
+static void idct32_1d(int16_t *input, int16_t *output) {
   int16_t step1[32], step2[32];
   int temp1, temp2;
 
--- a/vp9/common/vp9_invtrans.c
+++ b/vp9/common/vp9_invtrans.c
@@ -25,8 +25,7 @@
   for (i = 0; i < 16; i++) {
     TX_TYPE tx_type = get_tx_type_4x4(xd, &xd->block[i]);
     if (tx_type != DCT_DCT) {
-      vp9_short_iht4x4(xd->block[i].dqcoeff, xd->block[i].diff,
-                       32, tx_type);
+      vp9_short_iht4x4(xd->block[i].dqcoeff, xd->block[i].diff, 16, tx_type);
     } else {
       vp9_inverse_transform_b_4x4(xd, i, 32);
     }
@@ -58,8 +57,7 @@
   for (i = 0; i < 9; i += 8) {
     TX_TYPE tx_type = get_tx_type_8x8(xd, &xd->block[i]);
     if (tx_type != DCT_DCT) {
-      vp9_short_iht8x8(xd->block[i].dqcoeff, xd->block[i].diff,
-                           32, tx_type);
+      vp9_short_iht8x8(xd->block[i].dqcoeff, xd->block[i].diff, 16, tx_type);
     } else {
       vp9_inverse_transform_b_8x8(&blockd[i].dqcoeff[0],
                                   &blockd[i].diff[0], 32);
@@ -69,7 +67,7 @@
     TX_TYPE tx_type = get_tx_type_8x8(xd, &xd->block[i]);
     if (tx_type != DCT_DCT) {
       vp9_short_iht8x8(xd->block[i + 2].dqcoeff, xd->block[i].diff,
-                           32, tx_type);
+                           16, tx_type);
     } else {
       vp9_inverse_transform_b_8x8(&blockd[i + 2].dqcoeff[0],
                                   &blockd[i].diff[0], 32);
@@ -101,7 +99,7 @@
   BLOCKD *bd = &xd->block[0];
   TX_TYPE tx_type = get_tx_type_16x16(xd, bd);
   if (tx_type != DCT_DCT) {
-    vp9_short_iht16x16(bd->dqcoeff, bd->diff, 32, tx_type);
+    vp9_short_iht16x16(bd->dqcoeff, bd->diff, 16, tx_type);
   } else {
     vp9_inverse_transform_b_16x16(&xd->block[0].dqcoeff[0],
                                   &xd->block[0].diff[0], 32);
--- a/vp9/decoder/vp9_dequantize.c
+++ b/vp9/decoder/vp9_dequantize.c
@@ -65,7 +65,7 @@
   for (i = 0; i < 16; i++)
     input[i] = dq[i] * input[i];
 
-  vp9_short_iht4x4(input, output, 8, tx_type);
+  vp9_short_iht4x4(input, output, 4, tx_type);
   vpx_memset(input, 0, 32);
 
   add_residual(diff_ptr, pred, pitch, dest, stride, 4, 4);
@@ -86,7 +86,7 @@
     for (i = 1; i < 64; i++)
       input[i] *= dq[1];
 
-    vp9_short_iht8x8(input, output, 16, tx_type);
+    vp9_short_iht8x8(input, output, 8, tx_type);
     vpx_memset(input, 0, 128);
 
     add_residual(diff_ptr, pred, pitch, dest, stride, 8, 8);
@@ -247,7 +247,7 @@
       input[i] *= dq[1];
 
     // inverse hybrid transform
-    vp9_short_iht16x16(input, output, 32, tx_type);
+    vp9_short_iht16x16(input, output, 16, tx_type);
 
     // the idct halves ( >> 1) the pitch
     // vp9_short_idct16x16_c(input, output, 32);
--- a/vp9/encoder/vp9_dct.c
+++ b/vp9/encoder/vp9_dct.c
@@ -105,7 +105,6 @@
                         int pitch, TX_TYPE tx_type) {
   int16_t out[4 * 4];
   int16_t *outptr = &out[0];
-  const int short_pitch = pitch >> 1;
   int i, j;
   int16_t temp_in[4], temp_out[4];
 
@@ -137,7 +136,7 @@
   // column transform
   for (i = 0; i < 4; ++i) {
     for (j = 0; j < 4; ++j)
-      temp_in[j] = input[j * short_pitch + i] << 4;
+      temp_in[j] = input[j * pitch + i] << 4;
     if (i == 0 && temp_in[0])
       temp_in[0] += 1;
     fwdc(temp_in, temp_out);
@@ -308,7 +307,6 @@
                         int pitch, TX_TYPE tx_type) {
   int16_t out[64];
   int16_t *outptr = &out[0];
-  const int short_pitch = pitch >> 1;
   int i, j;
   int16_t temp_in[8], temp_out[8];
 
@@ -339,7 +337,7 @@
   // column transform
   for (i = 0; i < 8; ++i) {
     for (j = 0; j < 8; ++j)
-      temp_in[j] = input[j * short_pitch + i] << 2;
+      temp_in[j] = input[j * pitch + i] << 2;
     fwdc(temp_in, temp_out);
     for (j = 0; j < 8; ++j)
       outptr[j * 8 + i] = temp_out[j];
@@ -697,7 +695,6 @@
                           int pitch, TX_TYPE tx_type) {
   int16_t out[256];
   int16_t *outptr = &out[0];
-  const int short_pitch = pitch >> 1;
   int i, j;
   int16_t temp_in[16], temp_out[16];
 
@@ -728,7 +725,7 @@
   // column transform
   for (i = 0; i < 16; ++i) {
     for (j = 0; j < 16; ++j)
-      temp_in[j] = input[j * short_pitch + i] << 2;
+      temp_in[j] = input[j * pitch + i] << 2;
     fwdc(temp_in, temp_out);
     for (j = 0; j < 16; ++j)
       outptr[j * 16 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
--- a/vp9/encoder/vp9_encodeintra.c
+++ b/vp9/encoder/vp9_encodeintra.c
@@ -54,9 +54,9 @@
 
   tx_type = get_tx_type_4x4(&x->e_mbd, b);
   if (tx_type != DCT_DCT) {
-    vp9_short_fht4x4(be->src_diff, be->coeff, 32, tx_type);
+    vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type);
     vp9_ht_quantize_b_4x4(be, b, tx_type);
-    vp9_short_iht4x4(b->dqcoeff, b->diff, 32, tx_type);
+    vp9_short_iht4x4(b->dqcoeff, b->diff, 16, tx_type);
   } else {
     x->fwd_txm4x4(be->src_diff, be->coeff, 32);
     x->quantize_b_4x4(be, b) ;
@@ -149,10 +149,10 @@
 
     tx_type = get_tx_type_8x8(xd, &xd->block[ib]);
     if (tx_type != DCT_DCT) {
-      vp9_short_fht8x8(be->src_diff, (x->block + idx)->coeff, 32, tx_type);
+      vp9_short_fht8x8(be->src_diff, (x->block + idx)->coeff, 16, tx_type);
       x->quantize_b_8x8(x->block + idx, xd->block + idx);
       vp9_short_iht8x8(xd->block[idx].dqcoeff, xd->block[ib].diff,
-                            32, tx_type);
+                            16, tx_type);
     } else {
       x->fwd_txm8x8(be->src_diff, (x->block + idx)->coeff, 32);
       x->quantize_b_8x8(x->block + idx, xd->block + idx);
@@ -164,9 +164,9 @@
       be = &x->block[ib + iblock[i]];
       tx_type = get_tx_type_4x4(xd, b);
       if (tx_type != DCT_DCT) {
-        vp9_short_fht4x4(be->src_diff, be->coeff, 32, tx_type);
+        vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type);
         vp9_ht_quantize_b_4x4(be, b, tx_type);
-        vp9_short_iht4x4(b->dqcoeff, b->diff, 32, tx_type);
+        vp9_short_iht4x4(b->dqcoeff, b->diff, 16, tx_type);
       } else if (!(i & 1) && get_tx_type_4x4(xd, b + 1) == DCT_DCT) {
         x->fwd_txm8x4(be->src_diff, be->coeff, 32);
         x->quantize_b_4x4_pair(be, be + 1, b, b + 1);
--- a/vp9/encoder/vp9_encodemb.c
+++ b/vp9/encoder/vp9_encodemb.c
@@ -174,7 +174,7 @@
     BLOCK *b = &x->block[i];
     TX_TYPE tx_type = get_tx_type_4x4(xd, &xd->block[i]);
     if (tx_type != DCT_DCT) {
-      vp9_short_fht4x4(b->src_diff, b->coeff, 32, tx_type);
+      vp9_short_fht4x4(b->src_diff, b->coeff, 16, tx_type);
     } else if (!(i & 1) && get_tx_type_4x4(xd, &xd->block[i + 1]) == DCT_DCT) {
       x->fwd_txm8x4(&x->block[i].src_diff[0],
                            &x->block[i].coeff[0], 32);
@@ -209,7 +209,7 @@
     BLOCK *b = &x->block[i];
     tx_type = get_tx_type_8x8(xd, &xd->block[i]);
     if (tx_type != DCT_DCT) {
-      vp9_short_fht8x8(b->src_diff, b->coeff, 32, tx_type);
+      vp9_short_fht8x8(b->src_diff, b->coeff, 16, tx_type);
     } else {
       x->fwd_txm8x8(&x->block[i].src_diff[0],
                            &x->block[i].coeff[0], 32);
@@ -219,7 +219,7 @@
     BLOCK *b = &x->block[i];
     tx_type = get_tx_type_8x8(xd, &xd->block[i]);
     if (tx_type != DCT_DCT) {
-      vp9_short_fht8x8(b->src_diff, (b + 2)->coeff, 32, tx_type);
+      vp9_short_fht8x8(b->src_diff, (b + 2)->coeff, 16, tx_type);
     } else {
       x->fwd_txm8x8(&x->block[i].src_diff[0],
                            &x->block[i + 2].coeff[0], 32);
@@ -247,7 +247,7 @@
   TX_TYPE tx_type = get_tx_type_16x16(xd, &xd->block[0]);
   vp9_clear_system_state();
   if (tx_type != DCT_DCT) {
-    vp9_short_fht16x16(b->src_diff, b->coeff, 32, tx_type);
+    vp9_short_fht16x16(b->src_diff, b->coeff, 16, tx_type);
   } else {
     x->fwd_txm16x16(&x->block[0].src_diff[0],
                            &x->block[0].coeff[0], 32);
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -1013,7 +1013,7 @@
     b->bmi.as_mode.first = mode;
     tx_type = get_tx_type_4x4(xd, b);
     if (tx_type != DCT_DCT) {
-      vp9_short_fht4x4(be->src_diff, be->coeff, 32, tx_type);
+      vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type);
       vp9_ht_quantize_b_4x4(be, b, tx_type);
     } else {
       x->fwd_txm4x4(be->src_diff, be->coeff, 32);
@@ -1046,7 +1046,7 @@
 
   // inverse transform
   if (best_tx_type != DCT_DCT)
-    vp9_short_iht4x4(best_dqcoeff, b->diff, 32, best_tx_type);
+    vp9_short_iht4x4(best_dqcoeff, b->diff, 16, best_tx_type);
   else
     xd->inv_txm4x4(best_dqcoeff, b->diff, 32);
 
@@ -1309,7 +1309,7 @@
     if (xd->mode_info_context->mbmi.txfm_size == TX_8X8) {
       TX_TYPE tx_type = get_tx_type_8x8(xd, b);
       if (tx_type != DCT_DCT)
-        vp9_short_fht8x8(be->src_diff, (x->block + idx)->coeff, 32, tx_type);
+        vp9_short_fht8x8(be->src_diff, (x->block + idx)->coeff, 16, tx_type);
       else
         x->fwd_txm8x8(be->src_diff, (x->block + idx)->coeff, 32);
       x->quantize_b_8x8(x->block + idx, xd->block + idx);
@@ -1342,7 +1342,7 @@
         be = &x->block[ib + iblock[i]];
         tx_type = get_tx_type_4x4(xd, b);
         if (tx_type != DCT_DCT) {
-          vp9_short_fht4x4(be->src_diff, be->coeff, 32, tx_type);
+          vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type);
           vp9_ht_quantize_b_4x4(be, b, tx_type);
         } else if (!(i & 1) && get_tx_type_4x4(xd, b + 1) == DCT_DCT) {
           x->fwd_txm8x4(be->src_diff, be->coeff, 32);