libwebp: update to v0.5.1

third_party/libwebp/dsp/dec_sse2.c

 // Copyright 2011 Google Inc. All Rights Reserved.
 //
 // Use of this source code is governed by a BSD-style license
 // that can be found in the COPYING 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.
 // -----------------------------------------------------------------------------
 //
 // SSE2 version of some decoding functions (idct, loop filtering).
 //
 // Author: somnath@google.com (Somnath Banerjee)
 //         cduvivier@google.com (Christian Duvivier)
 
 #include "./dsp.h"
 
 #if defined(WEBP_USE_SSE2)
 
 // The 3-coeff sparse transform in SSE2 is not really faster than the plain-C
 // one it seems => disable it by default. Uncomment the following to enable:
 // #define USE_TRANSFORM_AC3
 
 #include <emmintrin.h>
+#include "./common_sse2.h"
 #include "../dec/vp8i.h"
+#include "../utils/utils.h"
 
 //------------------------------------------------------------------------------
 // Transforms (Paragraph 14.4)
 
 static void Transform(const int16_t* in, uint8_t* dst, int do_two) {
   // This implementation makes use of 16-bit fixed point versions of two
   // multiply constants:
   //    K1 = sqrt(2) * cos (pi/8) ~= 85627 / 2^16
   //    K2 = sqrt(2) * sin (pi/8) ~= 35468 / 2^16
   //
@@ skipping 60 matching lines @@
     const __m128i d4 = _mm_add_epi16(d1, d2);
     const __m128i d = _mm_add_epi16(d3, d4);
 
     // Second pass.
     const __m128i tmp0 = _mm_add_epi16(a, d);
     const __m128i tmp1 = _mm_add_epi16(b, c);
     const __m128i tmp2 = _mm_sub_epi16(b, c);
     const __m128i tmp3 = _mm_sub_epi16(a, d);
 
     // Transpose the two 4x4.
-    // a00 a01 a02 a03   b00 b01 b02 b03
-    // a10 a11 a12 a13   b10 b11 b12 b13
-    // a20 a21 a22 a23   b20 b21 b22 b23
-    // a30 a31 a32 a33   b30 b31 b32 b33
-    const __m128i transpose0_0 = _mm_unpacklo_epi16(tmp0, tmp1);
-    const __m128i transpose0_1 = _mm_unpacklo_epi16(tmp2, tmp3);
-    const __m128i transpose0_2 = _mm_unpackhi_epi16(tmp0, tmp1);
-    const __m128i transpose0_3 = _mm_unpackhi_epi16(tmp2, tmp3);
-    // a00 a10 a01 a11   a02 a12 a03 a13
-    // a20 a30 a21 a31   a22 a32 a23 a33
-    // b00 b10 b01 b11   b02 b12 b03 b13
-    // b20 b30 b21 b31   b22 b32 b23 b33
-    const __m128i transpose1_0 = _mm_unpacklo_epi32(transpose0_0, transpose0_1);
-    const __m128i transpose1_1 = _mm_unpacklo_epi32(transpose0_2, transpose0_3);
-    const __m128i transpose1_2 = _mm_unpackhi_epi32(transpose0_0, transpose0_1);
-    const __m128i transpose1_3 = _mm_unpackhi_epi32(transpose0_2, transpose0_3);
-    // a00 a10 a20 a30 a01 a11 a21 a31
-    // b00 b10 b20 b30 b01 b11 b21 b31
-    // a02 a12 a22 a32 a03 a13 a23 a33
-    // b02 b12 a22 b32 b03 b13 b23 b33
-    T0 = _mm_unpacklo_epi64(transpose1_0, transpose1_1);
-    T1 = _mm_unpackhi_epi64(transpose1_0, transpose1_1);
-    T2 = _mm_unpacklo_epi64(transpose1_2, transpose1_3);
-    T3 = _mm_unpackhi_epi64(transpose1_2, transpose1_3);
-    // a00 a10 a20 a30   b00 b10 b20 b30
-    // a01 a11 a21 a31   b01 b11 b21 b31
-    // a02 a12 a22 a32   b02 b12 b22 b32
-    // a03 a13 a23 a33   b03 b13 b23 b33
+    VP8Transpose_2_4x4_16b(&tmp0, &tmp1, &tmp2, &tmp3, &T0, &T1, &T2, &T3);
   }
 
   // Horizontal pass and subsequent transpose.
   {
     // First pass, c and d calculations are longer because of the "trick"
     // multiplications.
     const __m128i four = _mm_set1_epi16(4);
     const __m128i dc = _mm_add_epi16(T0, four);
     const __m128i a =  _mm_add_epi16(dc, T2);
     const __m128i b =  _mm_sub_epi16(dc, T2);
@@ skipping 14 matching lines @@
     const __m128i tmp0 = _mm_add_epi16(a, d);
     const __m128i tmp1 = _mm_add_epi16(b, c);
     const __m128i tmp2 = _mm_sub_epi16(b, c);
     const __m128i tmp3 = _mm_sub_epi16(a, d);
     const __m128i shifted0 = _mm_srai_epi16(tmp0, 3);
     const __m128i shifted1 = _mm_srai_epi16(tmp1, 3);
     const __m128i shifted2 = _mm_srai_epi16(tmp2, 3);
     const __m128i shifted3 = _mm_srai_epi16(tmp3, 3);
 
     // Transpose the two 4x4.
-    // a00 a01 a02 a03   b00 b01 b02 b03
-    // a10 a11 a12 a13   b10 b11 b12 b13
-    // a20 a21 a22 a23   b20 b21 b22 b23
-    // a30 a31 a32 a33   b30 b31 b32 b33
-    const __m128i transpose0_0 = _mm_unpacklo_epi16(shifted0, shifted1);
-    const __m128i transpose0_1 = _mm_unpacklo_epi16(shifted2, shifted3);
-    const __m128i transpose0_2 = _mm_unpackhi_epi16(shifted0, shifted1);
-    const __m128i transpose0_3 = _mm_unpackhi_epi16(shifted2, shifted3);
-    // a00 a10 a01 a11   a02 a12 a03 a13
-    // a20 a30 a21 a31   a22 a32 a23 a33
-    // b00 b10 b01 b11   b02 b12 b03 b13
-    // b20 b30 b21 b31   b22 b32 b23 b33
-    const __m128i transpose1_0 = _mm_unpacklo_epi32(transpose0_0, transpose0_1);
-    const __m128i transpose1_1 = _mm_unpacklo_epi32(transpose0_2, transpose0_3);
-    const __m128i transpose1_2 = _mm_unpackhi_epi32(transpose0_0, transpose0_1);
-    const __m128i transpose1_3 = _mm_unpackhi_epi32(transpose0_2, transpose0_3);
-    // a00 a10 a20 a30 a01 a11 a21 a31
-    // b00 b10 b20 b30 b01 b11 b21 b31
-    // a02 a12 a22 a32 a03 a13 a23 a33
-    // b02 b12 a22 b32 b03 b13 b23 b33
-    T0 = _mm_unpacklo_epi64(transpose1_0, transpose1_1);
-    T1 = _mm_unpackhi_epi64(transpose1_0, transpose1_1);
-    T2 = _mm_unpacklo_epi64(transpose1_2, transpose1_3);
-    T3 = _mm_unpackhi_epi64(transpose1_2, transpose1_3);
-    // a00 a10 a20 a30   b00 b10 b20 b30
-    // a01 a11 a21 a31   b01 b11 b21 b31
-    // a02 a12 a22 a32   b02 b12 b22 b32
-    // a03 a13 a23 a33   b03 b13 b23 b33
+    VP8Transpose_2_4x4_16b(&shifted0, &shifted1, &shifted2, &shifted3, &T0, &T1,
+                        &T2, &T3);
   }
 
   // Add inverse transform to 'dst' and store.
   {
     const __m128i zero = _mm_setzero_si128();
     // Load the reference(s).
     __m128i dst0, dst1, dst2, dst3;
     if (do_two) {
       // Load eight bytes/pixels per line.
       dst0 = _mm_loadl_epi64((__m128i*)(dst + 0 * BPS));
@@ skipping 1068 matching lines @@
   VP8PredChroma8[4] = DC8uvNoTop;
   VP8PredChroma8[5] = DC8uvNoLeft;
   VP8PredChroma8[6] = DC8uvNoTopLeft;
 }
 
 #else  // !WEBP_USE_SSE2
 
 WEBP_DSP_INIT_STUB(VP8DspInitSSE2)
 
 #endif  // WEBP_USE_SSE2