libvpx: Pull from upstream

source/libvpx/vp9/encoder/x86/vp9_dct_sse2.c

 /*
  *  Copyright (c) 2012 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.
  */
 
@@ skipping 224 matching lines @@
   u[0] = _mm_srai_epi32(v[0], DCT_CONST_BITS);
   u[1] = _mm_srai_epi32(v[1], DCT_CONST_BITS);
   u[2] = _mm_srai_epi32(v[2], DCT_CONST_BITS);
   u[3] = _mm_srai_epi32(v[3], DCT_CONST_BITS);
 
   in[0] = _mm_packs_epi32(u[0], u[2]);
   in[1] = _mm_packs_epi32(u[1], u[3]);
   transpose_4x4(in);
 }
 
-void vp9_short_fht4x4_sse2(const int16_t *input, int16_t *output,
-                           int stride, int tx_type) {
+void vp9_fht4x4_sse2(const int16_t *input, int16_t *output,
+                     int stride, int tx_type) {
   __m128i in[4];
-  load_buffer_4x4(input, in, stride);
+
   switch (tx_type) {
-    case 0:  // DCT_DCT
-      fdct4_sse2(in);
-      fdct4_sse2(in);
+    case DCT_DCT:
+      vp9_fdct4x4_sse2(input, output, stride);
       break;
-    case 1:  // ADST_DCT
+    case ADST_DCT:
+      load_buffer_4x4(input, in, stride);
       fadst4_sse2(in);
       fdct4_sse2(in);
+      write_buffer_4x4(output, in);
       break;
-    case 2:  // DCT_ADST
+    case DCT_ADST:
+      load_buffer_4x4(input, in, stride);
       fdct4_sse2(in);
       fadst4_sse2(in);
+      write_buffer_4x4(output, in);
       break;
-    case 3:  // ADST_ADST
+    case ADST_ADST:
+      load_buffer_4x4(input, in, stride);
       fadst4_sse2(in);
       fadst4_sse2(in);
+      write_buffer_4x4(output, in);
       break;
-    default:
-      assert(0);
-      break;
+   default:
+     assert(0);
+     break;
   }
-  write_buffer_4x4(output, in);
 }
 
 void vp9_fdct8x8_sse2(const int16_t *input, int16_t *output, int stride) {
   int pass;
   // Constants
   //    When we use them, in one case, they are all the same. In all others
   //    it's a pair of them that we need to repeat four times. This is done
   //    by constructing the 32 bit constant corresponding to that pair.
   const __m128i k__cospi_p16_p16 = _mm_set1_epi16(cospi_16_64);
   const __m128i k__cospi_p16_m16 = pair_set_epi16(cospi_16_64, -cospi_16_64);
@@ skipping 738 matching lines @@
   in[3] = _mm_sub_epi16(k__const_0, s2);
   in[4] = s3;
   in[5] = _mm_sub_epi16(k__const_0, s7);
   in[6] = s5;
   in[7] = _mm_sub_epi16(k__const_0, s1);
 
   // transpose
   array_transpose_8x8(in, in);
 }
 
-void vp9_short_fht8x8_sse2(const int16_t *input, int16_t *output,
-                           int stride, int tx_type) {
+void vp9_fht8x8_sse2(const int16_t *input, int16_t *output,
+                     int stride, int tx_type) {
   __m128i in[8];
-  load_buffer_8x8(input, in, stride);
+
   switch (tx_type) {
-    case 0:  // DCT_DCT
-      fdct8_sse2(in);
-      fdct8_sse2(in);
+    case DCT_DCT:
+      vp9_fdct8x8_sse2(input, output, stride);
       break;
-    case 1:  // ADST_DCT
+    case ADST_DCT:
+      load_buffer_8x8(input, in, stride);
       fadst8_sse2(in);
       fdct8_sse2(in);
+      right_shift_8x8(in, 1);
+      write_buffer_8x8(output, in, 8);
       break;
-    case 2:  // DCT_ADST
+    case DCT_ADST:
+      load_buffer_8x8(input, in, stride);
       fdct8_sse2(in);
       fadst8_sse2(in);
+      right_shift_8x8(in, 1);
+      write_buffer_8x8(output, in, 8);
       break;
-    case 3:  // ADST_ADST
+    case ADST_ADST:
+      load_buffer_8x8(input, in, stride);
       fadst8_sse2(in);
       fadst8_sse2(in);
+      right_shift_8x8(in, 1);
+      write_buffer_8x8(output, in, 8);
       break;
     default:
       assert(0);
       break;
   }
-  right_shift_8x8(in, 1);
-  write_buffer_8x8(output, in, 8);
 }
 
 void vp9_fdct16x16_sse2(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).
   int pass;
@@ skipping 1459 matching lines @@
   fdct16_8col(in1);
   array_transpose_16x16(in0, in1);
 }
 
 void fadst16_sse2(__m128i *in0, __m128i *in1) {
   fadst16_8col(in0);
   fadst16_8col(in1);
   array_transpose_16x16(in0, in1);
 }
 
-void vp9_short_fht16x16_sse2(const int16_t *input, int16_t *output,
-                             int stride, int tx_type) {
+void vp9_fht16x16_sse2(const int16_t *input, int16_t *output,
+                       int stride, int tx_type) {
   __m128i in0[16], in1[16];
-  load_buffer_16x16(input, in0, in1, stride);
+
   switch (tx_type) {
-    case 0:  // DCT_DCT
-      fdct16_sse2(in0, in1);
-      right_shift_16x16(in0, in1);
-      fdct16_sse2(in0, in1);
+    case DCT_DCT:
+      vp9_fdct16x16_sse2(input, output, stride);
       break;
-    case 1:  // ADST_DCT
+    case ADST_DCT:
+      load_buffer_16x16(input, in0, in1, stride);
       fadst16_sse2(in0, in1);
       right_shift_16x16(in0, in1);
       fdct16_sse2(in0, in1);
+      write_buffer_16x16(output, in0, in1, 16);
       break;
-    case 2:  // DCT_ADST
+    case DCT_ADST:
+      load_buffer_16x16(input, in0, in1, stride);
       fdct16_sse2(in0, in1);
       right_shift_16x16(in0, in1);
       fadst16_sse2(in0, in1);
+      write_buffer_16x16(output, in0, in1, 16);
       break;
-    case 3:  // ADST_ADST
+    case ADST_ADST:
+      load_buffer_16x16(input, in0, in1, stride);
       fadst16_sse2(in0, in1);
       right_shift_16x16(in0, in1);
       fadst16_sse2(in0, in1);
+      write_buffer_16x16(output, in0, in1, 16);
       break;
     default:
       assert(0);
       break;
   }
-  write_buffer_16x16(output, in0, in1, 16);
 }
 
 #define FDCT32x32_2D vp9_fdct32x32_rd_sse2
 #define FDCT32x32_HIGH_PRECISION 0
 #include "vp9/encoder/x86/vp9_dct32x32_sse2.c"
 #undef  FDCT32x32_2D
 #undef  FDCT32x32_HIGH_PRECISION
 
 #define FDCT32x32_2D vp9_fdct32x32_sse2
 #define FDCT32x32_HIGH_PRECISION 1
 #include "vp9/encoder/x86/vp9_dct32x32_sse2.c" // NOLINT
 #undef  FDCT32x32_2D
 #undef  FDCT32x32_HIGH_PRECISION