[x265] [PATCH] pixel8.inc: Optimization with sad_x4 4xn

praveen at multicorewareinc.com praveen at multicorewareinc.com
Mon Sep 2 09:10:26 CEST 2013


# HG changeset patch
# User praveen Tiwari
# Date 1378105815 -19800
# Node ID f42c5db55c77e7ea197c323e41c69c237fb00f21
# Parent  1a632b2e0fc7a3256202d3afd7dc187042081607
pixel8.inc: Optimization with sad_x4 4xn

diff -r 1a632b2e0fc7 -r f42c5db55c77 source/common/vec/pixel8.inc
--- a/source/common/vec/pixel8.inc	Mon Sep 02 11:58:04 2013 +0530
+++ b/source/common/vec/pixel8.inc	Mon Sep 02 12:40:15 2013 +0530
@@ -4377,634 +4377,644 @@
 {
     assert((ly % 4) == 0);
 
-    __m64 sum0 = _mm_setzero_si64();
-    __m64 sum1 = _mm_setzero_si64();
-    __m64 sum2 = _mm_setzero_si64();
-    __m64 sum3 = _mm_setzero_si64();
-
-    __m64 T00, T01, T02, T03, T04, T05, T06, T07;
-    __m64 T0, T1, T2, T3, T4, T5, T6, T7;
-    __m64 T10, T11, T12, T13, T14, T15, T16, T17;
-    __m64 T20, T21, T22, T23, T24, T25, T26, T27;
-
     if (ly == 4)
     {
-        T00 = _mm_cvtsi32_si64(*(int*)(fenc + 0 * FENC_STRIDE));
-        T01 = _mm_cvtsi32_si64(*(int*)(fenc + 1 * FENC_STRIDE));
-        T02 = _mm_cvtsi32_si64(*(int*)(fenc + 2 * FENC_STRIDE));
+        __m64 sum0, sum1, sum2, sum3;
+        __m64 T10, T11, T12, T13;
+        __m64 T00, T01, T02, T03;
+        __m64 T20, T21;
+
+        T00 = _mm_cvtsi32_si64(*(int*)(fenc));
+        T01 = _mm_cvtsi32_si64(*(int*)(fenc + FENC_STRIDE));
+        T00 = _mm_unpacklo_pi8(T00, T01);
+        T02 = _mm_cvtsi32_si64(*(int*)(fenc + (FENC_STRIDE << 1)));
         T03 = _mm_cvtsi32_si64(*(int*)(fenc + 3 * FENC_STRIDE));
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref1 + 0 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref1 + 1 * frefstride));
+        T02 = _mm_unpacklo_pi8(T02, T03);
+
+        T10 = _mm_cvtsi32_si64(*(int*)(fref1));
+        T11 = _mm_cvtsi32_si64(*(int*)(fref1 + frefstride));
+        T10 = _mm_unpacklo_pi8(T10, T11);
         T12 = _mm_cvtsi32_si64(*(int*)(fref1 + 2 * frefstride));
         T13 = _mm_cvtsi32_si64(*(int*)(fref1 + 3 * frefstride));
+        T12 = _mm_unpacklo_pi8(T12, T13);
 
         T20 = _mm_sad_pu8(T00, T10);
-        T21 = _mm_sad_pu8(T01, T11);
-        T22 = _mm_sad_pu8(T02, T12);
-        T23 = _mm_sad_pu8(T03, T13);
-
-        sum0 = _mm_add_pi16(sum0, T20);
-        sum0 = _mm_add_pi16(sum0, T21);
-        sum0 = _mm_add_pi16(sum0, T22);
-        sum0 = _mm_add_pi16(sum0, T23);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref2 + 0 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref2 + 1 * frefstride));
+        T21 = _mm_sad_pu8(T02, T12);
+
+        sum0 = _mm_add_pi16(T20, T21);
+
+        T10 = _mm_cvtsi32_si64(*(int*)(fref2));
+        T11 = _mm_cvtsi32_si64(*(int*)(fref2 + frefstride));
+        T10 = _mm_unpacklo_pi8(T10, T11);
         T12 = _mm_cvtsi32_si64(*(int*)(fref2 + 2 * frefstride));
         T13 = _mm_cvtsi32_si64(*(int*)(fref2 + 3 * frefstride));
+        T12 = _mm_unpacklo_pi8(T12, T13);
 
         T20 = _mm_sad_pu8(T00, T10);
-        T21 = _mm_sad_pu8(T01, T11);
-        T22 = _mm_sad_pu8(T02, T12);
-        T23 = _mm_sad_pu8(T03, T13);
-
-        sum1 = _mm_add_pi16(sum1, T20);
-        sum1 = _mm_add_pi16(sum1, T21);
-        sum1 = _mm_add_pi16(sum1, T22);
-        sum1 = _mm_add_pi16(sum1, T23);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref3 + 0 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref3 + 1 * frefstride));
+        T21 = _mm_sad_pu8(T02, T12);
+
+        sum1 = _mm_add_pi16(T20, T21);
+
+        T10 = _mm_cvtsi32_si64(*(int*)(fref3));
+        T11 = _mm_cvtsi32_si64(*(int*)(fref3 + frefstride));
+        T10 = _mm_unpacklo_pi8(T10, T11);
         T12 = _mm_cvtsi32_si64(*(int*)(fref3 + 2 * frefstride));
         T13 = _mm_cvtsi32_si64(*(int*)(fref3 + 3 * frefstride));
+        T12 = _mm_unpacklo_pi8(T12, T13);
 
         T20 = _mm_sad_pu8(T00, T10);
-        T21 = _mm_sad_pu8(T01, T11);
-        T22 = _mm_sad_pu8(T02, T12);
-        T23 = _mm_sad_pu8(T03, T13);
-
-        sum2 = _mm_add_pi16(sum2, T20);
-        sum2 = _mm_add_pi16(sum2, T21);
-        sum2 = _mm_add_pi16(sum2, T22);
-        sum2 = _mm_add_pi16(sum2, T23);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref4 + 0 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref4 + 1 * frefstride));
+        T21 = _mm_sad_pu8(T02, T12);
+
+        sum2 = _mm_add_pi16(T20, T21);
+
+        T10 = _mm_cvtsi32_si64(*(int*)(fref4));
+        T11 = _mm_cvtsi32_si64(*(int*)(fref4 + frefstride));
+        T10 = _mm_unpacklo_pi8(T10, T11);
         T12 = _mm_cvtsi32_si64(*(int*)(fref4 + 2 * frefstride));
         T13 = _mm_cvtsi32_si64(*(int*)(fref4 + 3 * frefstride));
+        T12 = _mm_unpacklo_pi8(T12, T13);
 
         T20 = _mm_sad_pu8(T00, T10);
-        T21 = _mm_sad_pu8(T01, T11);
-        T22 = _mm_sad_pu8(T02, T12);
-        T23 = _mm_sad_pu8(T03, T13);
-
-        sum3 = _mm_add_pi16(sum3, T20);
-        sum3 = _mm_add_pi16(sum3, T21);
-        sum3 = _mm_add_pi16(sum3, T22);
-        sum3 = _mm_add_pi16(sum3, T23);
+        T21 = _mm_sad_pu8(T02, T12);
+
+        sum3 = _mm_add_pi16(T20, T21);
+
+        res[0] = _m_to_int(sum0);
+        res[1] = _m_to_int(sum1);
+        res[2] = _m_to_int(sum2);
+        res[3] = _m_to_int(sum3);
     }
     else if (ly == 8)
     {
-        T00 = _mm_cvtsi32_si64(*(int*)(fenc + 0 * FENC_STRIDE));
-        T01 = _mm_cvtsi32_si64(*(int*)(fenc + 1 * FENC_STRIDE));
-        T02 = _mm_cvtsi32_si64(*(int*)(fenc + 2 * FENC_STRIDE));
-        T03 = _mm_cvtsi32_si64(*(int*)(fenc + 3 * FENC_STRIDE));
-        T04 = _mm_cvtsi32_si64(*(int*)(fenc + 4 * FENC_STRIDE));
-        T05 = _mm_cvtsi32_si64(*(int*)(fenc + 5 * FENC_STRIDE));
-        T06 = _mm_cvtsi32_si64(*(int*)(fenc + 6 * FENC_STRIDE));
-        T07 = _mm_cvtsi32_si64(*(int*)(fenc + 7 * FENC_STRIDE));
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref1 + 0 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref1 + 1 * frefstride));
-        T12 = _mm_cvtsi32_si64(*(int*)(fref1 + 2 * frefstride));
-        T13 = _mm_cvtsi32_si64(*(int*)(fref1 + 3 * frefstride));
-        T14 = _mm_cvtsi32_si64(*(int*)(fref1 + 4 * frefstride));
-        T15 = _mm_cvtsi32_si64(*(int*)(fref1 + 5 * frefstride));
-        T16 = _mm_cvtsi32_si64(*(int*)(fref1 + 6 * frefstride));
-        T17 = _mm_cvtsi32_si64(*(int*)(fref1 + 7 * frefstride));
-
-        T20 = _mm_sad_pu8(T00, T10);
-        T21 = _mm_sad_pu8(T01, T11);
-        T22 = _mm_sad_pu8(T02, T12);
-        T23 = _mm_sad_pu8(T03, T13);
-        T24 = _mm_sad_pu8(T04, T14);
-        T25 = _mm_sad_pu8(T05, T15);
-        T26 = _mm_sad_pu8(T06, T16);
-        T27 = _mm_sad_pu8(T07, T17);
-
-        sum0 = _mm_add_pi16(sum0, T20);
-        sum0 = _mm_add_pi16(sum0, T21);
-        sum0 = _mm_add_pi16(sum0, T22);
-        sum0 = _mm_add_pi16(sum0, T23);
-        sum0 = _mm_add_pi16(sum0, T24);
-        sum0 = _mm_add_pi16(sum0, T25);
-        sum0 = _mm_add_pi16(sum0, T26);
-        sum0 = _mm_add_pi16(sum0, T27);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref2 + 0 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref2 + 1 * frefstride));
-        T12 = _mm_cvtsi32_si64(*(int*)(fref2 + 2 * frefstride));
-        T13 = _mm_cvtsi32_si64(*(int*)(fref2 + 3 * frefstride));
-        T14 = _mm_cvtsi32_si64(*(int*)(fref2 + 4 * frefstride));
-        T15 = _mm_cvtsi32_si64(*(int*)(fref2 + 5 * frefstride));
-        T16 = _mm_cvtsi32_si64(*(int*)(fref2 + 6 * frefstride));
-        T17 = _mm_cvtsi32_si64(*(int*)(fref2 + 7 * frefstride));
-
-        T20 = _mm_sad_pu8(T00, T10);
-        T21 = _mm_sad_pu8(T01, T11);
-        T22 = _mm_sad_pu8(T02, T12);
-        T23 = _mm_sad_pu8(T03, T13);
-        T24 = _mm_sad_pu8(T04, T14);
-        T25 = _mm_sad_pu8(T05, T15);
-        T26 = _mm_sad_pu8(T06, T16);
-        T27 = _mm_sad_pu8(T07, T17);
-
-        sum1 = _mm_add_pi16(sum1, T20);
-        sum1 = _mm_add_pi16(sum1, T21);
-        sum1 = _mm_add_pi16(sum1, T22);
-        sum1 = _mm_add_pi16(sum1, T23);
-        sum1 = _mm_add_pi16(sum1, T24);
-        sum1 = _mm_add_pi16(sum1, T25);
-        sum1 = _mm_add_pi16(sum1, T26);
-        sum1 = _mm_add_pi16(sum1, T27);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref3 + 0 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref3 + 1 * frefstride));
-        T12 = _mm_cvtsi32_si64(*(int*)(fref3 + 2 * frefstride));
-        T13 = _mm_cvtsi32_si64(*(int*)(fref3 + 3 * frefstride));
-        T14 = _mm_cvtsi32_si64(*(int*)(fref3 + 4 * frefstride));
-        T15 = _mm_cvtsi32_si64(*(int*)(fref3 + 5 * frefstride));
-        T16 = _mm_cvtsi32_si64(*(int*)(fref3 + 6 * frefstride));
-        T17 = _mm_cvtsi32_si64(*(int*)(fref3 + 7 * frefstride));
-
-        T20 = _mm_sad_pu8(T00, T10);
-        T21 = _mm_sad_pu8(T01, T11);
-        T22 = _mm_sad_pu8(T02, T12);
-        T23 = _mm_sad_pu8(T03, T13);
-        T24 = _mm_sad_pu8(T04, T14);
-        T25 = _mm_sad_pu8(T05, T15);
-        T26 = _mm_sad_pu8(T06, T16);
-        T27 = _mm_sad_pu8(T07, T17);
-
-        sum2 = _mm_add_pi16(sum2, T20);
-        sum2 = _mm_add_pi16(sum2, T21);
-        sum2 = _mm_add_pi16(sum2, T22);
-        sum2 = _mm_add_pi16(sum2, T23);
-        sum2 = _mm_add_pi16(sum2, T24);
-        sum2 = _mm_add_pi16(sum2, T25);
-        sum2 = _mm_add_pi16(sum2, T26);
-        sum2 = _mm_add_pi16(sum2, T27);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref4 + 0 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref4 + 1 * frefstride));
-        T12 = _mm_cvtsi32_si64(*(int*)(fref4 + 2 * frefstride));
-        T13 = _mm_cvtsi32_si64(*(int*)(fref4 + 3 * frefstride));
-        T14 = _mm_cvtsi32_si64(*(int*)(fref4 + 4 * frefstride));
-        T15 = _mm_cvtsi32_si64(*(int*)(fref4 + 5 * frefstride));
-        T16 = _mm_cvtsi32_si64(*(int*)(fref4 + 6 * frefstride));
-        T17 = _mm_cvtsi32_si64(*(int*)(fref4 + 7 * frefstride));
-
-        T20 = _mm_sad_pu8(T00, T10);
-        T21 = _mm_sad_pu8(T01, T11);
-        T22 = _mm_sad_pu8(T02, T12);
-        T23 = _mm_sad_pu8(T03, T13);
-        T24 = _mm_sad_pu8(T04, T14);
-        T25 = _mm_sad_pu8(T05, T15);
-        T26 = _mm_sad_pu8(T06, T16);
-        T27 = _mm_sad_pu8(T07, T17);
-
-        sum3 = _mm_add_pi16(sum3, T20);
-        sum3 = _mm_add_pi16(sum3, T21);
-        sum3 = _mm_add_pi16(sum3, T22);
-        sum3 = _mm_add_pi16(sum3, T23);
-        sum3 = _mm_add_pi16(sum3, T24);
-        sum3 = _mm_add_pi16(sum3, T25);
-        sum3 = _mm_add_pi16(sum3, T26);
-        sum3 = _mm_add_pi16(sum3, T27);
+        __m128i sum0, sum1, sum2, sum3;
+
+        __m128i T00, T01, T02, T03;
+        __m128i T10, T11, T12, T13;
+        __m128i R00, R01, R02, R03, R04;
+        __m128i T20;
+
+        T00 = _mm_loadl_epi64((__m128i*)(fenc));
+        T01 = _mm_loadl_epi64((__m128i*)(fenc + FENC_STRIDE));
+        T01 = _mm_unpacklo_epi32(T00, T01);
+        T02 = _mm_loadl_epi64((__m128i*)(fenc + (2) * FENC_STRIDE));
+        T03 = _mm_loadl_epi64((__m128i*)(fenc + (3) * FENC_STRIDE));
+        T03 = _mm_unpacklo_epi32(T02, T03);
+        R00 = _mm_unpacklo_epi64(T01, T03);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref1));
+        T11 = _mm_loadl_epi64((__m128i*)(fref1 + frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref1 + 2 * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref1 + 3 * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R01 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref2));
+        T11 = _mm_loadl_epi64((__m128i*)(fref2 + frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref2 + 2 * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref2 + 3 * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R02 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref3));
+        T11 = _mm_loadl_epi64((__m128i*)(fref3 + frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref3 + 2 * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref3 + 3 * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R03 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref4));
+        T11 = _mm_loadl_epi64((__m128i*)(fref4 + frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref4 + 2 * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref4 + 3 * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R04 = _mm_unpacklo_epi64(T11, T13);
+
+        T20 = _mm_sad_epu8(R00, R01);
+        sum0 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+
+        T20 = _mm_sad_epu8(R00, R02);
+        sum1 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+
+        T20 = _mm_sad_epu8(R00, R03);
+        sum2 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+
+        T20 = _mm_sad_epu8(R00, R04);
+        sum3 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+
+        T00 = _mm_loadl_epi64((__m128i*)(fenc + (4) * FENC_STRIDE));
+        T01 = _mm_loadl_epi64((__m128i*)(fenc + (5) * FENC_STRIDE));
+        T01 = _mm_unpacklo_epi32(T00, T01);
+        T02 = _mm_loadl_epi64((__m128i*)(fenc + (6) * FENC_STRIDE));
+        T03 = _mm_loadl_epi64((__m128i*)(fenc + (7) * FENC_STRIDE));
+        T03 = _mm_unpacklo_epi32(T02, T03);
+        R00 = _mm_unpacklo_epi64(T01, T03);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref1 + 4 * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref1 + 5 * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref1 + 6 * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref1 + 7 * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R01 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref2 + 4 * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref2 + 5 * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref2 + 6 * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref2 + 7 * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R02 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref3 + 4 * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref3 + 5 * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref3 + 6 * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref3 + 7 * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R03 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref4 + 4 * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref4 + 5 * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref4 + 6 * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref4 + 7 * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R04 = _mm_unpacklo_epi64(T11, T13);
+
+        T20 = _mm_sad_epu8(R00, R01);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum0 = _mm_add_epi32(sum0, T20);
+
+        T20 = _mm_sad_epu8(R00, R02);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum1 = _mm_add_epi32(sum1, T20);
+
+        T20 = _mm_sad_epu8(R00, R03);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum2 = _mm_add_epi32(sum2, T20);
+
+        T20 = _mm_sad_epu8(R00, R04);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum3 = _mm_add_epi32(sum3, T20);
+
+        res[0] = _mm_cvtsi128_si32(sum0);
+        res[1] = _mm_cvtsi128_si32(sum1);
+        res[2] = _mm_cvtsi128_si32(sum2);
+        res[3] = _mm_cvtsi128_si32(sum3);
     }
     else if (ly == 16)
     {
-        T00 = _mm_cvtsi32_si64(*(int*)(fenc + 0 * FENC_STRIDE));
-        T01 = _mm_cvtsi32_si64(*(int*)(fenc + 1 * FENC_STRIDE));
-        T02 = _mm_cvtsi32_si64(*(int*)(fenc + 2 * FENC_STRIDE));
-        T03 = _mm_cvtsi32_si64(*(int*)(fenc + 3 * FENC_STRIDE));
-        T04 = _mm_cvtsi32_si64(*(int*)(fenc + 4 * FENC_STRIDE));
-        T05 = _mm_cvtsi32_si64(*(int*)(fenc + 5 * FENC_STRIDE));
-        T06 = _mm_cvtsi32_si64(*(int*)(fenc + 6 * FENC_STRIDE));
-        T07 = _mm_cvtsi32_si64(*(int*)(fenc + 7 * FENC_STRIDE));
-        T0 = _mm_cvtsi32_si64(*(int*)(fenc +  8 * FENC_STRIDE));
-        T1 = _mm_cvtsi32_si64(*(int*)(fenc +  9 * FENC_STRIDE));
-        T2 = _mm_cvtsi32_si64(*(int*)(fenc +  10 * FENC_STRIDE));
-        T3 = _mm_cvtsi32_si64(*(int*)(fenc +  11 * FENC_STRIDE));
-        T4 = _mm_cvtsi32_si64(*(int*)(fenc + 12 * FENC_STRIDE));
-        T5 = _mm_cvtsi32_si64(*(int*)(fenc + 13 * FENC_STRIDE));
-        T6 = _mm_cvtsi32_si64(*(int*)(fenc + 14 * FENC_STRIDE));
-        T7 = _mm_cvtsi32_si64(*(int*)(fenc + 15 * FENC_STRIDE));
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref1 + 0 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref1 + 1 * frefstride));
-        T12 = _mm_cvtsi32_si64(*(int*)(fref1 + 2 * frefstride));
-        T13 = _mm_cvtsi32_si64(*(int*)(fref1 + 3 * frefstride));
-        T14 = _mm_cvtsi32_si64(*(int*)(fref1 + 4 * frefstride));
-        T15 = _mm_cvtsi32_si64(*(int*)(fref1 + 5 * frefstride));
-        T16 = _mm_cvtsi32_si64(*(int*)(fref1 + 6 * frefstride));
-        T17 = _mm_cvtsi32_si64(*(int*)(fref1 + 7 * frefstride));
-
-        T20 = _mm_sad_pu8(T00, T10);
-        T21 = _mm_sad_pu8(T01, T11);
-        T22 = _mm_sad_pu8(T02, T12);
-        T23 = _mm_sad_pu8(T03, T13);
-        T24 = _mm_sad_pu8(T04, T14);
-        T25 = _mm_sad_pu8(T05, T15);
-        T26 = _mm_sad_pu8(T06, T16);
-        T27 = _mm_sad_pu8(T07, T17);
-
-        sum0 = _mm_add_pi16(sum0, T20);
-        sum0 = _mm_add_pi16(sum0, T21);
-        sum0 = _mm_add_pi16(sum0, T22);
-        sum0 = _mm_add_pi16(sum0, T23);
-        sum0 = _mm_add_pi16(sum0, T24);
-        sum0 = _mm_add_pi16(sum0, T25);
-        sum0 = _mm_add_pi16(sum0, T26);
-        sum0 = _mm_add_pi16(sum0, T27);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref1 + 8 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref1 + 9 * frefstride));
-        T12 = _mm_cvtsi32_si64(*(int*)(fref1 + 10 * frefstride));
-        T13 = _mm_cvtsi32_si64(*(int*)(fref1 + 11 * frefstride));
-        T14 = _mm_cvtsi32_si64(*(int*)(fref1 + 12 * frefstride));
-        T15 = _mm_cvtsi32_si64(*(int*)(fref1 + 13 * frefstride));
-        T16 = _mm_cvtsi32_si64(*(int*)(fref1 + 14 * frefstride));
-        T17 = _mm_cvtsi32_si64(*(int*)(fref1 + 15 * frefstride));
-
-        T20 = _mm_sad_pu8(T0, T10);
-        T21 = _mm_sad_pu8(T1, T11);
-        T22 = _mm_sad_pu8(T2, T12);
-        T23 = _mm_sad_pu8(T3, T13);
-        T24 = _mm_sad_pu8(T4, T14);
-        T25 = _mm_sad_pu8(T5, T15);
-        T26 = _mm_sad_pu8(T6, T16);
-        T27 = _mm_sad_pu8(T7, T17);
-
-        sum0 = _mm_add_pi16(sum0, T20);
-        sum0 = _mm_add_pi16(sum0, T21);
-        sum0 = _mm_add_pi16(sum0, T22);
-        sum0 = _mm_add_pi16(sum0, T23);
-        sum0 = _mm_add_pi16(sum0, T24);
-        sum0 = _mm_add_pi16(sum0, T25);
-        sum0 = _mm_add_pi16(sum0, T26);
-        sum0 = _mm_add_pi16(sum0, T27);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref2 + 0 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref2 + 1 * frefstride));
-        T12 = _mm_cvtsi32_si64(*(int*)(fref2 + 2 * frefstride));
-        T13 = _mm_cvtsi32_si64(*(int*)(fref2 + 3 * frefstride));
-        T14 = _mm_cvtsi32_si64(*(int*)(fref2 + 4 * frefstride));
-        T15 = _mm_cvtsi32_si64(*(int*)(fref2 + 5 * frefstride));
-        T16 = _mm_cvtsi32_si64(*(int*)(fref2 + 6 * frefstride));
-        T17 = _mm_cvtsi32_si64(*(int*)(fref2 + 7 * frefstride));
-
-        T20 = _mm_sad_pu8(T00, T10);
-        T21 = _mm_sad_pu8(T01, T11);
-        T22 = _mm_sad_pu8(T02, T12);
-        T23 = _mm_sad_pu8(T03, T13);
-        T24 = _mm_sad_pu8(T04, T14);
-        T25 = _mm_sad_pu8(T05, T15);
-        T26 = _mm_sad_pu8(T06, T16);
-        T27 = _mm_sad_pu8(T07, T17);
-
-        sum1 = _mm_add_pi16(sum1, T20);
-        sum1 = _mm_add_pi16(sum1, T21);
-        sum1 = _mm_add_pi16(sum1, T22);
-        sum1 = _mm_add_pi16(sum1, T23);
-        sum1 = _mm_add_pi16(sum1, T24);
-        sum1 = _mm_add_pi16(sum1, T25);
-        sum1 = _mm_add_pi16(sum1, T26);
-        sum1 = _mm_add_pi16(sum1, T27);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref2 + 8 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref2 + 9 * frefstride));
-        T12 = _mm_cvtsi32_si64(*(int*)(fref2 + 10 * frefstride));
-        T13 = _mm_cvtsi32_si64(*(int*)(fref2 + 11 * frefstride));
-        T14 = _mm_cvtsi32_si64(*(int*)(fref2 + 12 * frefstride));
-        T15 = _mm_cvtsi32_si64(*(int*)(fref2 + 13 * frefstride));
-        T16 = _mm_cvtsi32_si64(*(int*)(fref2 + 14 * frefstride));
-        T17 = _mm_cvtsi32_si64(*(int*)(fref2 + 15 * frefstride));
-
-        T20 = _mm_sad_pu8(T0, T10);
-        T21 = _mm_sad_pu8(T1, T11);
-        T22 = _mm_sad_pu8(T2, T12);
-        T23 = _mm_sad_pu8(T3, T13);
-        T24 = _mm_sad_pu8(T4, T14);
-        T25 = _mm_sad_pu8(T5, T15);
-        T26 = _mm_sad_pu8(T6, T16);
-        T27 = _mm_sad_pu8(T7, T17);
-
-        sum1 = _mm_add_pi16(sum1, T20);
-        sum1 = _mm_add_pi16(sum1, T21);
-        sum1 = _mm_add_pi16(sum1, T22);
-        sum1 = _mm_add_pi16(sum1, T23);
-        sum1 = _mm_add_pi16(sum1, T24);
-        sum1 = _mm_add_pi16(sum1, T25);
-        sum1 = _mm_add_pi16(sum1, T26);
-        sum1 = _mm_add_pi16(sum1, T27);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref3 + 0 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref3 + 1 * frefstride));
-        T12 = _mm_cvtsi32_si64(*(int*)(fref3 + 2 * frefstride));
-        T13 = _mm_cvtsi32_si64(*(int*)(fref3 + 3 * frefstride));
-        T14 = _mm_cvtsi32_si64(*(int*)(fref3 + 4 * frefstride));
-        T15 = _mm_cvtsi32_si64(*(int*)(fref3 + 5 * frefstride));
-        T16 = _mm_cvtsi32_si64(*(int*)(fref3 + 6 * frefstride));
-        T17 = _mm_cvtsi32_si64(*(int*)(fref3 + 7 * frefstride));
-
-        T20 = _mm_sad_pu8(T00, T10);
-        T21 = _mm_sad_pu8(T01, T11);
-        T22 = _mm_sad_pu8(T02, T12);
-        T23 = _mm_sad_pu8(T03, T13);
-        T24 = _mm_sad_pu8(T04, T14);
-        T25 = _mm_sad_pu8(T05, T15);
-        T26 = _mm_sad_pu8(T06, T16);
-        T27 = _mm_sad_pu8(T07, T17);
-
-        sum2 = _mm_add_pi16(sum2, T20);
-        sum2 = _mm_add_pi16(sum2, T21);
-        sum2 = _mm_add_pi16(sum2, T22);
-        sum2 = _mm_add_pi16(sum2, T23);
-        sum2 = _mm_add_pi16(sum2, T24);
-        sum2 = _mm_add_pi16(sum2, T25);
-        sum2 = _mm_add_pi16(sum2, T26);
-        sum2 = _mm_add_pi16(sum2, T27);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref3 + 8 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref3 + 9 * frefstride));
-        T12 = _mm_cvtsi32_si64(*(int*)(fref3 + 10 * frefstride));
-        T13 = _mm_cvtsi32_si64(*(int*)(fref3 + 11 * frefstride));
-        T14 = _mm_cvtsi32_si64(*(int*)(fref3 + 12 * frefstride));
-        T15 = _mm_cvtsi32_si64(*(int*)(fref3 + 13 * frefstride));
-        T16 = _mm_cvtsi32_si64(*(int*)(fref3 + 14 * frefstride));
-        T17 = _mm_cvtsi32_si64(*(int*)(fref3 + 15 * frefstride));
-
-        T20 = _mm_sad_pu8(T0, T10);
-        T21 = _mm_sad_pu8(T1, T11);
-        T22 = _mm_sad_pu8(T2, T12);
-        T23 = _mm_sad_pu8(T3, T13);
-        T24 = _mm_sad_pu8(T4, T14);
-        T25 = _mm_sad_pu8(T5, T15);
-        T26 = _mm_sad_pu8(T6, T16);
-        T27 = _mm_sad_pu8(T7, T17);
-
-        sum2 = _mm_add_pi16(sum2, T20);
-        sum2 = _mm_add_pi16(sum2, T21);
-        sum2 = _mm_add_pi16(sum2, T22);
-        sum2 = _mm_add_pi16(sum2, T23);
-        sum2 = _mm_add_pi16(sum2, T24);
-        sum2 = _mm_add_pi16(sum2, T25);
-        sum2 = _mm_add_pi16(sum2, T26);
-        sum2 = _mm_add_pi16(sum2, T27);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref4 + 0 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref4 + 1 * frefstride));
-        T12 = _mm_cvtsi32_si64(*(int*)(fref4 + 2 * frefstride));
-        T13 = _mm_cvtsi32_si64(*(int*)(fref4 + 3 * frefstride));
-        T14 = _mm_cvtsi32_si64(*(int*)(fref4 + 4 * frefstride));
-        T15 = _mm_cvtsi32_si64(*(int*)(fref4 + 5 * frefstride));
-        T16 = _mm_cvtsi32_si64(*(int*)(fref4 + 6 * frefstride));
-        T17 = _mm_cvtsi32_si64(*(int*)(fref4 + 7 * frefstride));
-
-        T20 = _mm_sad_pu8(T00, T10);
-        T21 = _mm_sad_pu8(T01, T11);
-        T22 = _mm_sad_pu8(T02, T12);
-        T23 = _mm_sad_pu8(T03, T13);
-        T24 = _mm_sad_pu8(T04, T14);
-        T25 = _mm_sad_pu8(T05, T15);
-        T26 = _mm_sad_pu8(T06, T16);
-        T27 = _mm_sad_pu8(T07, T17);
-
-        sum3 = _mm_add_pi16(sum3, T20);
-        sum3 = _mm_add_pi16(sum3, T21);
-        sum3 = _mm_add_pi16(sum3, T22);
-        sum3 = _mm_add_pi16(sum3, T23);
-        sum3 = _mm_add_pi16(sum3, T24);
-        sum3 = _mm_add_pi16(sum3, T25);
-        sum3 = _mm_add_pi16(sum3, T26);
-        sum3 = _mm_add_pi16(sum3, T27);
-
-        T10 = _mm_cvtsi32_si64(*(int*)(fref4 + 8 * frefstride));
-        T11 = _mm_cvtsi32_si64(*(int*)(fref4 + 9 * frefstride));
-        T12 = _mm_cvtsi32_si64(*(int*)(fref4 + 10 * frefstride));
-        T13 = _mm_cvtsi32_si64(*(int*)(fref4 + 11 * frefstride));
-        T14 = _mm_cvtsi32_si64(*(int*)(fref4 + 12 * frefstride));
-        T15 = _mm_cvtsi32_si64(*(int*)(fref4 + 13 * frefstride));
-        T16 = _mm_cvtsi32_si64(*(int*)(fref4 + 14 * frefstride));
-        T17 = _mm_cvtsi32_si64(*(int*)(fref4 + 15 * frefstride));
-
-        T20 = _mm_sad_pu8(T0, T10);
-        T21 = _mm_sad_pu8(T1, T11);
-        T22 = _mm_sad_pu8(T2, T12);
-        T23 = _mm_sad_pu8(T3, T13);
-        T24 = _mm_sad_pu8(T4, T14);
-        T25 = _mm_sad_pu8(T5, T15);
-        T26 = _mm_sad_pu8(T6, T16);
-        T27 = _mm_sad_pu8(T7, T17);
-
-        sum3 = _mm_add_pi16(sum3, T20);
-        sum3 = _mm_add_pi16(sum3, T21);
-        sum3 = _mm_add_pi16(sum3, T22);
-        sum3 = _mm_add_pi16(sum3, T23);
-        sum3 = _mm_add_pi16(sum3, T24);
-        sum3 = _mm_add_pi16(sum3, T25);
-        sum3 = _mm_add_pi16(sum3, T26);
-        sum3 = _mm_add_pi16(sum3, T27);
+        __m128i sum0, sum1, sum2, sum3;
+
+        __m128i T00, T01, T02, T03;
+        __m128i T10, T11, T12, T13;
+        __m128i R00, R01, R02, R03, R04;
+        __m128i T20;
+
+        T00 = _mm_loadl_epi64((__m128i*)(fenc));
+        T01 = _mm_loadl_epi64((__m128i*)(fenc + FENC_STRIDE));
+        T01 = _mm_unpacklo_epi32(T00, T01);
+        T02 = _mm_loadl_epi64((__m128i*)(fenc + (2) * FENC_STRIDE));
+        T03 = _mm_loadl_epi64((__m128i*)(fenc + (3) * FENC_STRIDE));
+        T03 = _mm_unpacklo_epi32(T02, T03);
+        R00 = _mm_unpacklo_epi64(T01, T03);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref1));
+        T11 = _mm_loadl_epi64((__m128i*)(fref1 + frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref1 + (2) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref1 + (3) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R01 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref2));
+        T11 = _mm_loadl_epi64((__m128i*)(fref2 + frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref2 + (2) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref2 + (3) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R02 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref3));
+        T11 = _mm_loadl_epi64((__m128i*)(fref3 + frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref3 + (2) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref3 + (3) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R03 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref4));
+        T11 = _mm_loadl_epi64((__m128i*)(fref4 + frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref4 + (2) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref4 + (3) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R04 = _mm_unpacklo_epi64(T11, T13);
+
+        T20 = _mm_sad_epu8(R00, R01);
+        sum0 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+
+        T20 = _mm_sad_epu8(R00, R02);
+        sum1 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+
+        T20 = _mm_sad_epu8(R00, R03);
+        sum2 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+
+        T20 = _mm_sad_epu8(R00, R04);
+        sum3 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+
+        T00 = _mm_loadl_epi64((__m128i*)(fenc + (4) * FENC_STRIDE));
+        T01 = _mm_loadl_epi64((__m128i*)(fenc + (5) * FENC_STRIDE));
+        T01 = _mm_unpacklo_epi32(T00, T01);
+        T02 = _mm_loadl_epi64((__m128i*)(fenc + (6) * FENC_STRIDE));
+        T03 = _mm_loadl_epi64((__m128i*)(fenc + (7) * FENC_STRIDE));
+        T03 = _mm_unpacklo_epi32(T02, T03);
+        R00 = _mm_unpacklo_epi64(T01, T03);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref1 + (4) * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref1 + (5) * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref1 + (6) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref1 + (7) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R01 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref2 + (4) * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref2 + (5) * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref2 + (6) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref2 + (7) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R02 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref3 + (4) * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref3 + (5) * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref3 + (6) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref3 + (7) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R03 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref4 + (4) * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref4 + (5) * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref4 + (6) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref4 + (7) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R04 = _mm_unpacklo_epi64(T11, T13);
+
+        T20 = _mm_sad_epu8(R00, R01);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum0 = _mm_add_epi32(sum0, T20);
+
+        T20 = _mm_sad_epu8(R00, R02);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum1 = _mm_add_epi32(sum1, T20);
+
+        T20 = _mm_sad_epu8(R00, R03);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum2 = _mm_add_epi32(sum2, T20);
+
+        T20 = _mm_sad_epu8(R00, R04);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum3 = _mm_add_epi32(sum3, T20);
+
+        T00 = _mm_loadl_epi64((__m128i*)(fenc + (8) * FENC_STRIDE));
+        T01 = _mm_loadl_epi64((__m128i*)(fenc + (9) * FENC_STRIDE));
+        T01 = _mm_unpacklo_epi32(T00, T01);
+        T02 = _mm_loadl_epi64((__m128i*)(fenc + (10) * FENC_STRIDE));
+        T03 = _mm_loadl_epi64((__m128i*)(fenc + (11) * FENC_STRIDE));
+        T03 = _mm_unpacklo_epi32(T02, T03);
+        R00 = _mm_unpacklo_epi64(T01, T03);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref1 + (8) * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref1 + (9) * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref1 + (10) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref1 + (11) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R01 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref2 + (8) * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref2 + (9) * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref2 + (10) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref2 + (11) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R02 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref3 + (8) * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref3 + (9) * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref3 + (10) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref3 + (11) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R03 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref4 + (8) * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref4 + (9) * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref4 + (10) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref4 + (11) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R04 = _mm_unpacklo_epi64(T11, T13);
+
+        T20 = _mm_sad_epu8(R00, R01);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum0 = _mm_add_epi32(sum0, T20);
+
+        T20 = _mm_sad_epu8(R00, R02);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum1 = _mm_add_epi32(sum1, T20);
+
+        T20 = _mm_sad_epu8(R00, R03);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum2 = _mm_add_epi32(sum2, T20);
+
+        T20 = _mm_sad_epu8(R00, R04);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum3 = _mm_add_epi32(sum3, T20);
+
+        T00 = _mm_loadl_epi64((__m128i*)(fenc + (12) * FENC_STRIDE));
+        T01 = _mm_loadl_epi64((__m128i*)(fenc + (13) * FENC_STRIDE));
+        T01 = _mm_unpacklo_epi32(T00, T01);
+        T02 = _mm_loadl_epi64((__m128i*)(fenc + (14) * FENC_STRIDE));
+        T03 = _mm_loadl_epi64((__m128i*)(fenc + (15) * FENC_STRIDE));
+        T03 = _mm_unpacklo_epi32(T02, T03);
+        R00 = _mm_unpacklo_epi64(T01, T03);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref1 + (12) * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref1 + (13) * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref1 + (14) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref1 + (15) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R01 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref2 + (12) * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref2 + (13) * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref2 + (14) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref2 + (15) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R02 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref3 + (12) * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref3 + (13) * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref3 + (14) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref3 + (15) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R03 = _mm_unpacklo_epi64(T11, T13);
+
+        T10 = _mm_loadl_epi64((__m128i*)(fref4 + (12) * frefstride));
+        T11 = _mm_loadl_epi64((__m128i*)(fref4 + (13) * frefstride));
+        T11 = _mm_unpacklo_epi32(T10, T11);
+        T12 = _mm_loadl_epi64((__m128i*)(fref4 + (14) * frefstride));
+        T13 = _mm_loadl_epi64((__m128i*)(fref4 + (15) * frefstride));
+        T13 = _mm_unpacklo_epi32(T12, T13);
+        R04 = _mm_unpacklo_epi64(T11, T13);
+
+        T20 = _mm_sad_epu8(R00, R01);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum0 = _mm_add_epi32(sum0, T20);
+
+        T20 = _mm_sad_epu8(R00, R02);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum1 = _mm_add_epi32(sum1, T20);
+
+        T20 = _mm_sad_epu8(R00, R03);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum2 = _mm_add_epi32(sum2, T20);
+
+        T20 = _mm_sad_epu8(R00, R04);
+        T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+        sum3 = _mm_add_epi32(sum3, T20);
+
+        res[0] = _mm_cvtsi128_si32(sum0);
+        res[1] = _mm_cvtsi128_si32(sum1);
+        res[2] = _mm_cvtsi128_si32(sum2);
+        res[3] = _mm_cvtsi128_si32(sum3);
     }
     else if ((ly % 8) == 0)
     {
+        __m128i sum0 = _mm_setzero_si128();
+        __m128i sum1 = _mm_setzero_si128();
+        __m128i sum2 = _mm_setzero_si128();
+        __m128i sum3 = _mm_setzero_si128();
+
+        __m128i T00, T01, T02, T03;
+        __m128i T10, T11, T12, T13;
+        __m128i R00, R01, R02, R03, R04;
+        __m128i T20;
+
         for (int i = 0; i < ly; i += 8)
         {
+            T00 = _mm_loadl_epi64((__m128i*)(fenc + (i + 0) * FENC_STRIDE));
+            T01 = _mm_loadl_epi64((__m128i*)(fenc + (i + 1) * FENC_STRIDE));
+            T01 = _mm_unpacklo_epi32(T00, T01);
+            T02 = _mm_loadl_epi64((__m128i*)(fenc + (i + 2) * FENC_STRIDE));
+            T03 = _mm_loadl_epi64((__m128i*)(fenc + (i + 3) * FENC_STRIDE));
+            T03 = _mm_unpacklo_epi32(T02, T03);
+            R00 = _mm_unpacklo_epi64(T01, T03);
+
+            T10 = _mm_loadl_epi64((__m128i*)(fref1 + (i + 0) * frefstride));
+            T11 = _mm_loadl_epi64((__m128i*)(fref1 + (i + 1) * frefstride));
+            T11 = _mm_unpacklo_epi32(T10, T11);
+            T12 = _mm_loadl_epi64((__m128i*)(fref1 + (i + 2) * frefstride));
+            T13 = _mm_loadl_epi64((__m128i*)(fref1 + (i + 3) * frefstride));
+            T13 = _mm_unpacklo_epi32(T12, T13);
+            R01 = _mm_unpacklo_epi64(T11, T13);
+
+            T10 = _mm_loadl_epi64((__m128i*)(fref2 + (i + 0) * frefstride));
+            T11 = _mm_loadl_epi64((__m128i*)(fref2 + (i + 1) * frefstride));
+            T11 = _mm_unpacklo_epi32(T10, T11);
+            T12 = _mm_loadl_epi64((__m128i*)(fref2 + (i + 2) * frefstride));
+            T13 = _mm_loadl_epi64((__m128i*)(fref2 + (i + 3) * frefstride));
+            T13 = _mm_unpacklo_epi32(T12, T13);
+            R02 = _mm_unpacklo_epi64(T11, T13);
+
+            T10 = _mm_loadl_epi64((__m128i*)(fref3 + (i + 0) * frefstride));
+            T11 = _mm_loadl_epi64((__m128i*)(fref3 + (i + 1) * frefstride));
+            T11 = _mm_unpacklo_epi32(T10, T11);
+            T12 = _mm_loadl_epi64((__m128i*)(fref3 + (i + 2) * frefstride));
+            T13 = _mm_loadl_epi64((__m128i*)(fref3 + (i + 3) * frefstride));
+            T13 = _mm_unpacklo_epi32(T12, T13);
+            R03 = _mm_unpacklo_epi64(T11, T13);
+
+            T10 = _mm_loadl_epi64((__m128i*)(fref4 + (i + 0) * frefstride));
+            T11 = _mm_loadl_epi64((__m128i*)(fref4 + (i + 1) * frefstride));
+            T11 = _mm_unpacklo_epi32(T10, T11);
+            T12 = _mm_loadl_epi64((__m128i*)(fref4 + (i + 2) * frefstride));
+            T13 = _mm_loadl_epi64((__m128i*)(fref4 + (i + 3) * frefstride));
+            T13 = _mm_unpacklo_epi32(T12, T13);
+            R04 = _mm_unpacklo_epi64(T11, T13);
+
+            T20 = _mm_sad_epu8(R00, R01);
+            T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+            sum0 = _mm_add_epi32(sum0, T20);
+
+            T20 = _mm_sad_epu8(R00, R02);
+            T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+            sum1 = _mm_add_epi32(sum1, T20);
+
+            T20 = _mm_sad_epu8(R00, R03);
+            T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+            sum2 = _mm_add_epi32(sum2, T20);
+
+            T20 = _mm_sad_epu8(R00, R04);
+            T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+            sum3 = _mm_add_epi32(sum3, T20);
+
+            T00 = _mm_loadl_epi64((__m128i*)(fenc + (i + 4) * FENC_STRIDE));
+            T01 = _mm_loadl_epi64((__m128i*)(fenc + (i + 5) * FENC_STRIDE));
+            T01 = _mm_unpacklo_epi32(T00, T01);
+            T02 = _mm_loadl_epi64((__m128i*)(fenc + (i + 6) * FENC_STRIDE));
+            T03 = _mm_loadl_epi64((__m128i*)(fenc + (i + 7) * FENC_STRIDE));
+            T03 = _mm_unpacklo_epi32(T02, T03);
+            R00 = _mm_unpacklo_epi64(T01, T03);
+
+            T10 = _mm_loadl_epi64((__m128i*)(fref1 + (i + 4) * frefstride));
+            T11 = _mm_loadl_epi64((__m128i*)(fref1 + (i + 5) * frefstride));
+            T11 = _mm_unpacklo_epi32(T10, T11);
+            T12 = _mm_loadl_epi64((__m128i*)(fref1 + (i + 6) * frefstride));
+            T13 = _mm_loadl_epi64((__m128i*)(fref1 + (i + 7) * frefstride));
+            T13 = _mm_unpacklo_epi32(T12, T13);
+            R01 = _mm_unpacklo_epi64(T11, T13);
+
+            T10 = _mm_loadl_epi64((__m128i*)(fref2 + (i + 4) * frefstride));
+            T11 = _mm_loadl_epi64((__m128i*)(fref2 + (i + 5) * frefstride));
+            T11 = _mm_unpacklo_epi32(T10, T11);
+            T12 = _mm_loadl_epi64((__m128i*)(fref2 + (i + 6) * frefstride));
+            T13 = _mm_loadl_epi64((__m128i*)(fref2 + (i + 7) * frefstride));
+            T13 = _mm_unpacklo_epi32(T12, T13);
+            R02 = _mm_unpacklo_epi64(T11, T13);
+
+            T10 = _mm_loadl_epi64((__m128i*)(fref3 + (i + 4) * frefstride));
+            T11 = _mm_loadl_epi64((__m128i*)(fref3 + (i + 5) * frefstride));
+            T11 = _mm_unpacklo_epi32(T10, T11);
+            T12 = _mm_loadl_epi64((__m128i*)(fref3 + (i + 6) * frefstride));
+            T13 = _mm_loadl_epi64((__m128i*)(fref3 + (i + 7) * frefstride));
+            T13 = _mm_unpacklo_epi32(T12, T13);
+            R03 = _mm_unpacklo_epi64(T11, T13);
+
+            T10 = _mm_loadl_epi64((__m128i*)(fref4 + (i + 4) * frefstride));
+            T11 = _mm_loadl_epi64((__m128i*)(fref4 + (i + 5) * frefstride));
+            T11 = _mm_unpacklo_epi32(T10, T11);
+            T12 = _mm_loadl_epi64((__m128i*)(fref4 + (i + 6) * frefstride));
+            T13 = _mm_loadl_epi64((__m128i*)(fref4 + (i + 7) * frefstride));
+            T13 = _mm_unpacklo_epi32(T12, T13);
+            R04 = _mm_unpacklo_epi64(T11, T13);
+
+            T20 = _mm_sad_epu8(R00, R01);
+            T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+            sum0 = _mm_add_epi32(sum0, T20);
+
+            T20 = _mm_sad_epu8(R00, R02);
+            T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+            sum1 = _mm_add_epi32(sum1, T20);
+
+            T20 = _mm_sad_epu8(R00, R03);
+            T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+            sum2 = _mm_add_epi32(sum2, T20);
+
+            T20 = _mm_sad_epu8(R00, R04);
+            T20 = _mm_add_epi32(T20, _mm_shuffle_epi32(T20, 2));
+            sum3 = _mm_add_epi32(sum3, T20);
+        }
+
+        res[0] = _mm_cvtsi128_si32(sum0);
+        res[1] = _mm_cvtsi128_si32(sum1);
+        res[2] = _mm_cvtsi128_si32(sum2);
+        res[3] = _mm_cvtsi128_si32(sum3);
+    }
+    else
+    {
+        __m64 sum0 = _mm_setzero_si64();
+        __m64 sum1 = _mm_setzero_si64();
+        __m64 sum2 = _mm_setzero_si64();
+        __m64 sum3 = _mm_setzero_si64();
+
+        __m64 T10, T11, T12, T13;
+        __m64 T00, T01, T02, T03;
+        __m64 T20, T21;
+
+        for (int i = 0; i < ly; i += 4)
+        {
+            int frefstrideZero = (i + 0) * frefstride;
+            int frefstrideOne = (i + 1) * frefstride;
+            int frefstrideTwo = (i + 2) * frefstride;
+            int frefstrideThree = (i + 3) * frefstride;
+
             T00 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 0) * FENC_STRIDE));
             T01 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 1) * FENC_STRIDE));
+            T00 = _mm_unpacklo_pi8(T00, T01);
             T02 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 2) * FENC_STRIDE));
             T03 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 3) * FENC_STRIDE));
-            T04 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 4) * FENC_STRIDE));
-            T05 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 5) * FENC_STRIDE));
-            T06 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 6) * FENC_STRIDE));
-            T07 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 7) * FENC_STRIDE));
-
-            T10 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 0) * frefstride));
-            T11 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 1) * frefstride));
-            T12 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 2) * frefstride));
-            T13 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 3) * frefstride));
-            T14 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 4) * frefstride));
-            T15 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 5) * frefstride));
-            T16 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 6) * frefstride));
-            T17 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 7) * frefstride));
+            T02 = _mm_unpacklo_pi8(T02, T03);
+
+            T10 = _mm_cvtsi32_si64(*(int*)(fref1 + frefstrideZero));
+            T11 = _mm_cvtsi32_si64(*(int*)(fref1 + frefstrideOne));
+            T10 = _mm_unpacklo_pi8(T10, T11);
+            T12 = _mm_cvtsi32_si64(*(int*)(fref1 + frefstrideTwo));
+            T13 = _mm_cvtsi32_si64(*(int*)(fref1 + frefstrideThree));
+            T12 = _mm_unpacklo_pi8(T12, T13);
 
             T20 = _mm_sad_pu8(T00, T10);
-            T21 = _mm_sad_pu8(T01, T11);
-            T22 = _mm_sad_pu8(T02, T12);
-            T23 = _mm_sad_pu8(T03, T13);
-            T24 = _mm_sad_pu8(T04, T14);
-            T25 = _mm_sad_pu8(T05, T15);
-            T26 = _mm_sad_pu8(T06, T16);
-            T27 = _mm_sad_pu8(T07, T17);
+            T21 = _mm_sad_pu8(T02, T12);
 
             sum0 = _mm_add_pi16(sum0, T20);
             sum0 = _mm_add_pi16(sum0, T21);
-            sum0 = _mm_add_pi16(sum0, T22);
-            sum0 = _mm_add_pi16(sum0, T23);
-            sum0 = _mm_add_pi16(sum0, T24);
-            sum0 = _mm_add_pi16(sum0, T25);
-            sum0 = _mm_add_pi16(sum0, T26);
-            sum0 = _mm_add_pi16(sum0, T27);
-
-            T10 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 0) * frefstride));
-            T11 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 1) * frefstride));
-            T12 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 2) * frefstride));
-            T13 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 3) * frefstride));
-            T14 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 4) * frefstride));
-            T15 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 5) * frefstride));
-            T16 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 6) * frefstride));
-            T17 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 7) * frefstride));
+
+            T10 = _mm_cvtsi32_si64(*(int*)(fref2 + frefstrideZero));
+            T11 = _mm_cvtsi32_si64(*(int*)(fref2 + frefstrideOne));
+            T10 = _mm_unpacklo_pi8(T10, T11);
+            T12 = _mm_cvtsi32_si64(*(int*)(fref2 + frefstrideTwo));
+            T13 = _mm_cvtsi32_si64(*(int*)(fref2 + frefstrideThree));
+            T12 = _mm_unpacklo_pi8(T12, T13);
 
             T20 = _mm_sad_pu8(T00, T10);
-            T21 = _mm_sad_pu8(T01, T11);
-            T22 = _mm_sad_pu8(T02, T12);
-            T23 = _mm_sad_pu8(T03, T13);
-            T24 = _mm_sad_pu8(T04, T14);
-            T25 = _mm_sad_pu8(T05, T15);
-            T26 = _mm_sad_pu8(T06, T16);
-            T27 = _mm_sad_pu8(T07, T17);
+            T21 = _mm_sad_pu8(T02, T12);
 
             sum1 = _mm_add_pi16(sum1, T20);
             sum1 = _mm_add_pi16(sum1, T21);
-            sum1 = _mm_add_pi16(sum1, T22);
-            sum1 = _mm_add_pi16(sum1, T23);
-            sum1 = _mm_add_pi16(sum1, T24);
-            sum1 = _mm_add_pi16(sum1, T25);
-            sum1 = _mm_add_pi16(sum1, T26);
-            sum1 = _mm_add_pi16(sum1, T27);
-
-            T10 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 0) * frefstride));
-            T11 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 1) * frefstride));
-            T12 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 2) * frefstride));
-            T13 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 3) * frefstride));
-            T14 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 4) * frefstride));
-            T15 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 5) * frefstride));
-            T16 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 6) * frefstride));
-            T17 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 7) * frefstride));
+
+            T10 = _mm_cvtsi32_si64(*(int*)(fref3 + frefstrideZero));
+            T11 = _mm_cvtsi32_si64(*(int*)(fref3 + frefstrideOne));
+            T10 = _mm_unpacklo_pi8(T10, T11);
+            T12 = _mm_cvtsi32_si64(*(int*)(fref3 + frefstrideTwo));
+            T13 = _mm_cvtsi32_si64(*(int*)(fref3 + frefstrideThree));
+            T12 = _mm_unpacklo_pi8(T12, T13);
 
             T20 = _mm_sad_pu8(T00, T10);
-            T21 = _mm_sad_pu8(T01, T11);
-            T22 = _mm_sad_pu8(T02, T12);
-            T23 = _mm_sad_pu8(T03, T13);
-            T24 = _mm_sad_pu8(T04, T14);
-            T25 = _mm_sad_pu8(T05, T15);
-            T26 = _mm_sad_pu8(T06, T16);
-            T27 = _mm_sad_pu8(T07, T17);
+            T21 = _mm_sad_pu8(T02, T12);
 
             sum2 = _mm_add_pi16(sum2, T20);
             sum2 = _mm_add_pi16(sum2, T21);
-            sum2 = _mm_add_pi16(sum2, T22);
-            sum2 = _mm_add_pi16(sum2, T23);
-            sum2 = _mm_add_pi16(sum2, T24);
-            sum2 = _mm_add_pi16(sum2, T25);
-            sum2 = _mm_add_pi16(sum2, T26);
-            sum2 = _mm_add_pi16(sum2, T27);
-
-            T10 = _mm_cvtsi32_si64(*(int*)(fref4 + (i + 0) * frefstride));
-            T11 = _mm_cvtsi32_si64(*(int*)(fref4 + (i + 1) * frefstride));
-            T12 = _mm_cvtsi32_si64(*(int*)(fref4 + (i + 2) * frefstride));
-            T13 = _mm_cvtsi32_si64(*(int*)(fref4 + (i + 3) * frefstride));
-            T14 = _mm_cvtsi32_si64(*(int*)(fref4 + (i + 4) * frefstride));
-            T15 = _mm_cvtsi32_si64(*(int*)(fref4 + (i + 5) * frefstride));
-            T16 = _mm_cvtsi32_si64(*(int*)(fref4 + (i + 6) * frefstride));
-            T17 = _mm_cvtsi32_si64(*(int*)(fref4 + (i + 7) * frefstride));
+
+            T10 = _mm_cvtsi32_si64(*(int*)(fref4 + frefstrideZero));
+            T11 = _mm_cvtsi32_si64(*(int*)(fref4 + frefstrideOne));
+            T10 = _mm_unpacklo_pi8(T10, T11);
+            T12 = _mm_cvtsi32_si64(*(int*)(fref4 + frefstrideTwo));
+            T13 = _mm_cvtsi32_si64(*(int*)(fref4 + frefstrideThree));
+            T12 = _mm_unpacklo_pi8(T12, T13);
 
             T20 = _mm_sad_pu8(T00, T10);
-            T21 = _mm_sad_pu8(T01, T11);
-            T22 = _mm_sad_pu8(T02, T12);
-            T23 = _mm_sad_pu8(T03, T13);
-            T24 = _mm_sad_pu8(T04, T14);
-            T25 = _mm_sad_pu8(T05, T15);
-            T26 = _mm_sad_pu8(T06, T16);
-            T27 = _mm_sad_pu8(T07, T17);
+            T21 = _mm_sad_pu8(T02, T12);
 
             sum3 = _mm_add_pi16(sum3, T20);
             sum3 = _mm_add_pi16(sum3, T21);
-            sum3 = _mm_add_pi16(sum3, T22);
-            sum3 = _mm_add_pi16(sum3, T23);
-            sum3 = _mm_add_pi16(sum3, T24);
-            sum3 = _mm_add_pi16(sum3, T25);
-            sum3 = _mm_add_pi16(sum3, T26);
-            sum3 = _mm_add_pi16(sum3, T27);
         }
-    }
-    else
-    {
-        for (int i = 0; i < ly; i += 4)
-        {
-            T00 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 0) * FENC_STRIDE));
-            T01 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 1) * FENC_STRIDE));
-            T02 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 2) * FENC_STRIDE));
-            T03 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 3) * FENC_STRIDE));
-
-            T10 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 0) * frefstride));
-            T11 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 1) * frefstride));
-            T12 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 2) * frefstride));
-            T13 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 3) * frefstride));
-
-            T20 = _mm_sad_pu8(T00, T10);
-            T21 = _mm_sad_pu8(T01, T11);
-            T22 = _mm_sad_pu8(T02, T12);
-            T23 = _mm_sad_pu8(T03, T13);
-
-            sum0 = _mm_add_pi16(sum0, T20);
-            sum0 = _mm_add_pi16(sum0, T21);
-            sum0 = _mm_add_pi16(sum0, T22);
-            sum0 = _mm_add_pi16(sum0, T23);
-
-            T10 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 0) * frefstride));
-            T11 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 1) * frefstride));
-            T12 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 2) * frefstride));
-            T13 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 3) * frefstride));
-
-            T20 = _mm_sad_pu8(T00, T10);
-            T21 = _mm_sad_pu8(T01, T11);
-            T22 = _mm_sad_pu8(T02, T12);
-            T23 = _mm_sad_pu8(T03, T13);
-
-            sum1 = _mm_add_pi16(sum1, T20);
-            sum1 = _mm_add_pi16(sum1, T21);
-            sum1 = _mm_add_pi16(sum1, T22);
-            sum1 = _mm_add_pi16(sum1, T23);
-
-            T10 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 0) * frefstride));
-            T11 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 1) * frefstride));
-            T12 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 2) * frefstride));
-            T13 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 3) * frefstride));
-
-            T20 = _mm_sad_pu8(T00, T10);
-            T21 = _mm_sad_pu8(T01, T11);
-            T22 = _mm_sad_pu8(T02, T12);
-            T23 = _mm_sad_pu8(T03, T13);
-
-            sum2 = _mm_add_pi16(sum2, T20);
-            sum2 = _mm_add_pi16(sum2, T21);
-            sum2 = _mm_add_pi16(sum2, T22);
-            sum2 = _mm_add_pi16(sum2, T23);
-
-            T10 = _mm_cvtsi32_si64(*(int*)(fref4 + (i + 0) * frefstride));
-            T11 = _mm_cvtsi32_si64(*(int*)(fref4 + (i + 1) * frefstride));
-            T12 = _mm_cvtsi32_si64(*(int*)(fref4 + (i + 2) * frefstride));
-            T13 = _mm_cvtsi32_si64(*(int*)(fref4 + (i + 3) * frefstride));
-
-            T20 = _mm_sad_pu8(T00, T10);
-            T21 = _mm_sad_pu8(T01, T11);
-            T22 = _mm_sad_pu8(T02, T12);
-            T23 = _mm_sad_pu8(T03, T13);
-
-            sum3 = _mm_add_pi16(sum3, T20);
-            sum3 = _mm_add_pi16(sum3, T21);
-            sum3 = _mm_add_pi16(sum3, T22);
-            sum3 = _mm_add_pi16(sum3, T23);
-        }
-    }
-
-    res[0] = _m_to_int(sum0);
-    res[1] = _m_to_int(sum1);
-    res[2] = _m_to_int(sum2);
-    res[3] = _m_to_int(sum3);
+
+        res[0] = _m_to_int(sum0);
+        res[1] = _m_to_int(sum1);
+        res[2] = _m_to_int(sum2);
+        res[3] = _m_to_int(sum3);
+    }
 }
 
 #else /* if HAVE_MMX */


More information about the x265-devel mailing list