[x265] [PATCH] pixel: simplify sad_x4_32 to make it easier to maintain

dnyaneshwar at multicorewareinc.com dnyaneshwar at multicorewareinc.com
Mon Oct 7 13:06:36 CEST 2013


# HG changeset patch
# User Dnyaneshwar Gorade <dnyaneshwar at multicorewareinc.com>
# Date 1381143965 -19800
#      Mon Oct 07 16:36:05 2013 +0530
# Node ID 981fa87b0839cb0b909603694034cce3b4ad1c3d
# Parent  211cf42529631d870de2140d41f0e73073265387
pixel: simplify sad_x4_32 to make it easier to maintain

diff -r 211cf4252963 -r 981fa87b0839 source/common/vec/pixel-sse41.cpp
--- a/source/common/vec/pixel-sse41.cpp	Mon Oct 07 16:29:42 2013 +0530
+++ b/source/common/vec/pixel-sse41.cpp	Mon Oct 07 16:36:05 2013 +0530
@@ -3567,304 +3567,129 @@
 void sad_x4_32(pixel *fenc, pixel *fref1, pixel *fref2, pixel *fref3, pixel *fref4, intptr_t frefstride, int *res)
 {
     res[0] = res[1] = res[2] = res[3] = 0;
+    __m128i T00, T01, T02, T03, T04, T05, T06, T07;
+    __m128i T10, T11, T12, T13, T14, T15, T16, T17;
+    __m128i T20, T21, T22, T23, T24, T25, T26, T27;
+    __m128i sum0, sum1;
+
+#define PROCESS_32x4x4(BASE) \
+        T00 = _mm_load_si128((__m128i*)(fenc + (BASE) * FENC_STRIDE)); \
+        T01 = _mm_load_si128((__m128i*)(fenc + (BASE + 1) * FENC_STRIDE)); \
+        T02 = _mm_load_si128((__m128i*)(fenc + (BASE + 2) * FENC_STRIDE)); \
+        T03 = _mm_load_si128((__m128i*)(fenc + (BASE + 3) * FENC_STRIDE)); \
+        T04 = _mm_load_si128((__m128i*)(fenc + (BASE) * FENC_STRIDE + 16)); \
+        T05 = _mm_load_si128((__m128i*)(fenc + (BASE + 1) * FENC_STRIDE + 16)); \
+        T06 = _mm_load_si128((__m128i*)(fenc + (BASE + 2) * FENC_STRIDE + 16)); \
+        T07 = _mm_load_si128((__m128i*)(fenc + (BASE + 3) * FENC_STRIDE + 16)); \
+        T10 = _mm_loadu_si128((__m128i*)(fref1 + (BASE) * frefstride)); \
+        T11 = _mm_loadu_si128((__m128i*)(fref1 + (BASE + 1) * frefstride)); \
+        T12 = _mm_loadu_si128((__m128i*)(fref1 + (BASE + 2) * frefstride)); \
+        T13 = _mm_loadu_si128((__m128i*)(fref1 + (BASE + 3) * frefstride)); \
+        T14 = _mm_loadu_si128((__m128i*)(fref1 + (BASE) * frefstride + 16)); \
+        T15 = _mm_loadu_si128((__m128i*)(fref1 + (BASE + 1) * frefstride + 16)); \
+        T16 = _mm_loadu_si128((__m128i*)(fref1 + (BASE + 2) * frefstride + 16)); \
+        T17 = _mm_loadu_si128((__m128i*)(fref1 + (BASE + 3) * frefstride + 16)); \
+        T20 = _mm_sad_epu8(T00, T10); \
+        T21 = _mm_sad_epu8(T01, T11); \
+        T22 = _mm_sad_epu8(T02, T12); \
+        T23 = _mm_sad_epu8(T03, T13); \
+        T24 = _mm_sad_epu8(T04, T14); \
+        T25 = _mm_sad_epu8(T05, T15); \
+        T26 = _mm_sad_epu8(T06, T16); \
+        T27 = _mm_sad_epu8(T07, T17); \
+        T20 = _mm_add_epi16(T20, T21); \
+        T22 = _mm_add_epi16(T22, T23); \
+        T24 = _mm_add_epi16(T24, T25); \
+        T26 = _mm_add_epi16(T26, T27); \
+        sum0 = _mm_add_epi16(T20, T22); \
+        sum0 = _mm_add_epi16(sum0, T24); \
+        sum0 = _mm_add_epi16(sum0, T26); \
+        sum1 = _mm_shuffle_epi32(sum0, 2); \
+        sum0 = _mm_add_epi32(sum0, sum1); \
+        res[0] += _mm_cvtsi128_si32(sum0); \
+        T10 = _mm_loadu_si128((__m128i*)(fref2 + (BASE) * frefstride)); \
+        T11 = _mm_loadu_si128((__m128i*)(fref2 + (BASE + 1) * frefstride)); \
+        T12 = _mm_loadu_si128((__m128i*)(fref2 + (BASE + 2) * frefstride)); \
+        T13 = _mm_loadu_si128((__m128i*)(fref2 + (BASE + 3) * frefstride)); \
+        T14 = _mm_loadu_si128((__m128i*)(fref2 + (BASE) * frefstride + 16)); \
+        T15 = _mm_loadu_si128((__m128i*)(fref2 + (BASE + 1) * frefstride + 16)); \
+        T16 = _mm_loadu_si128((__m128i*)(fref2 + (BASE + 2) * frefstride + 16)); \
+        T17 = _mm_loadu_si128((__m128i*)(fref2 + (BASE + 3) * frefstride + 16)); \
+        T20 = _mm_sad_epu8(T00, T10); \
+        T21 = _mm_sad_epu8(T01, T11); \
+        T22 = _mm_sad_epu8(T02, T12); \
+        T23 = _mm_sad_epu8(T03, T13); \
+        T24 = _mm_sad_epu8(T04, T14); \
+        T25 = _mm_sad_epu8(T05, T15); \
+        T26 = _mm_sad_epu8(T06, T16); \
+        T27 = _mm_sad_epu8(T07, T17); \
+        T20 = _mm_add_epi16(T20, T21); \
+        T22 = _mm_add_epi16(T22, T23); \
+        T24 = _mm_add_epi16(T24, T25); \
+        T26 = _mm_add_epi16(T26, T27); \
+        sum0 = _mm_add_epi16(T20, T22); \
+        sum0 = _mm_add_epi16(sum0, T24); \
+        sum0 = _mm_add_epi16(sum0, T26); \
+        sum1 = _mm_shuffle_epi32(sum0, 2); \
+        sum0 = _mm_add_epi32(sum0, sum1); \
+        res[1] += _mm_cvtsi128_si32(sum0); \
+        T10 = _mm_loadu_si128((__m128i*)(fref3 + (BASE) * frefstride)); \
+        T11 = _mm_loadu_si128((__m128i*)(fref3 + (BASE + 1) * frefstride)); \
+        T12 = _mm_loadu_si128((__m128i*)(fref3 + (BASE + 2) * frefstride)); \
+        T13 = _mm_loadu_si128((__m128i*)(fref3 + (BASE + 3) * frefstride)); \
+        T14 = _mm_loadu_si128((__m128i*)(fref3 + (BASE) * frefstride + 16)); \
+        T15 = _mm_loadu_si128((__m128i*)(fref3 + (BASE + 1) * frefstride + 16)); \
+        T16 = _mm_loadu_si128((__m128i*)(fref3 + (BASE + 2) * frefstride + 16)); \
+        T17 = _mm_loadu_si128((__m128i*)(fref3 + (BASE + 3) * frefstride + 16)); \
+        T20 = _mm_sad_epu8(T00, T10); \
+        T21 = _mm_sad_epu8(T01, T11); \
+        T22 = _mm_sad_epu8(T02, T12); \
+        T23 = _mm_sad_epu8(T03, T13); \
+        T24 = _mm_sad_epu8(T04, T14); \
+        T25 = _mm_sad_epu8(T05, T15); \
+        T26 = _mm_sad_epu8(T06, T16); \
+        T27 = _mm_sad_epu8(T07, T17); \
+        T20 = _mm_add_epi16(T20, T21); \
+        T22 = _mm_add_epi16(T22, T23); \
+        T24 = _mm_add_epi16(T24, T25); \
+        T26 = _mm_add_epi16(T26, T27); \
+        sum0 = _mm_add_epi16(T20, T22); \
+        sum0 = _mm_add_epi16(sum0, T24); \
+        sum0 = _mm_add_epi16(sum0, T26); \
+        sum1 = _mm_shuffle_epi32(sum0, 2); \
+        sum0 = _mm_add_epi32(sum0, sum1); \
+        res[2] += _mm_cvtsi128_si32(sum0); \
+        T10 = _mm_loadu_si128((__m128i*)(fref4 + (BASE) * frefstride)); \
+        T11 = _mm_loadu_si128((__m128i*)(fref4 + (BASE + 1) * frefstride)); \
+        T12 = _mm_loadu_si128((__m128i*)(fref4 + (BASE + 2) * frefstride)); \
+        T13 = _mm_loadu_si128((__m128i*)(fref4 + (BASE + 3) * frefstride)); \
+        T14 = _mm_loadu_si128((__m128i*)(fref4 + (BASE) * frefstride + 16)); \
+        T15 = _mm_loadu_si128((__m128i*)(fref4 + (BASE + 1) * frefstride + 16)); \
+        T16 = _mm_loadu_si128((__m128i*)(fref4 + (BASE + 2) * frefstride + 16)); \
+        T17 = _mm_loadu_si128((__m128i*)(fref4 + (BASE + 3) * frefstride + 16)); \
+        T20 = _mm_sad_epu8(T00, T10); \
+        T21 = _mm_sad_epu8(T01, T11); \
+        T22 = _mm_sad_epu8(T02, T12); \
+        T23 = _mm_sad_epu8(T03, T13); \
+        T24 = _mm_sad_epu8(T04, T14); \
+        T25 = _mm_sad_epu8(T05, T15); \
+        T26 = _mm_sad_epu8(T06, T16); \
+        T27 = _mm_sad_epu8(T07, T17); \
+        T20 = _mm_add_epi16(T20, T21); \
+        T22 = _mm_add_epi16(T22, T23); \
+        T24 = _mm_add_epi16(T24, T25); \
+        T26 = _mm_add_epi16(T26, T27); \
+        sum0 = _mm_add_epi16(T20, T22); \
+        sum0 = _mm_add_epi16(sum0, T24); \
+        sum0 = _mm_add_epi16(sum0, T26); \
+        sum1 = _mm_shuffle_epi32(sum0, 2); \
+        sum0 = _mm_add_epi32(sum0, sum1); \
+        res[3] += _mm_cvtsi128_si32(sum0)
+
     for(int i = 0; i < ly; i += 8)
     {
-        __m128i T00, T01, T02, T03, T04, T05, T06, T07;
-        __m128i T10, T11, T12, T13, T14, T15, T16, T17;
-        __m128i T20, T21, T22, T23, T24, T25, T26, T27;
-        __m128i sum0, sum1;
-
-        T00 = _mm_load_si128((__m128i*)(fenc + (i) * FENC_STRIDE));
-        T01 = _mm_load_si128((__m128i*)(fenc + (i + 1) * FENC_STRIDE));
-        T02 = _mm_load_si128((__m128i*)(fenc + (i + 2) * FENC_STRIDE));
-        T03 = _mm_load_si128((__m128i*)(fenc + (i + 3) * FENC_STRIDE));
-
-        T04 = _mm_load_si128((__m128i*)(fenc + (i) * FENC_STRIDE + 16));
-        T05 = _mm_load_si128((__m128i*)(fenc + (i + 1) * FENC_STRIDE + 16));
-        T06 = _mm_load_si128((__m128i*)(fenc + (i + 2) * FENC_STRIDE + 16));
-        T07 = _mm_load_si128((__m128i*)(fenc + (i + 3) * FENC_STRIDE + 16));
-
-        T10 = _mm_loadu_si128((__m128i*)(fref1 + (i) * frefstride));
-        T11 = _mm_loadu_si128((__m128i*)(fref1 + (i + 1) * frefstride));
-        T12 = _mm_loadu_si128((__m128i*)(fref1 + (i + 2) * frefstride));
-        T13 = _mm_loadu_si128((__m128i*)(fref1 + (i + 3) * frefstride));
-
-        T14 = _mm_loadu_si128((__m128i*)(fref1 + (i) * frefstride + 16));
-        T15 = _mm_loadu_si128((__m128i*)(fref1 + (i + 1) * frefstride + 16));
-        T16 = _mm_loadu_si128((__m128i*)(fref1 + (i + 2) * frefstride + 16));
-        T17 = _mm_loadu_si128((__m128i*)(fref1 + (i + 3) * frefstride + 16));
-
-        T20 = _mm_sad_epu8(T00, T10);
-        T21 = _mm_sad_epu8(T01, T11);
-        T22 = _mm_sad_epu8(T02, T12);
-        T23 = _mm_sad_epu8(T03, T13);
-
-        T24 = _mm_sad_epu8(T04, T14);
-        T25 = _mm_sad_epu8(T05, T15);
-        T26 = _mm_sad_epu8(T06, T16);
-        T27 = _mm_sad_epu8(T07, T17);
-
-        T20 = _mm_add_epi16(T20, T21);
-        T22 = _mm_add_epi16(T22, T23);
-
-        T24 = _mm_add_epi16(T24, T25);
-        T26 = _mm_add_epi16(T26, T27);
-        
-        sum0 = _mm_add_epi16(T20, T22);
-        sum0 = _mm_add_epi16(sum0, T24);
-        sum0 = _mm_add_epi16(sum0, T26);
-
-        sum1 = _mm_shuffle_epi32(sum0, 2);
-        sum0 = _mm_add_epi32(sum0, sum1);
-        res[0] += _mm_cvtsi128_si32(sum0);
-
-        T10 = _mm_loadu_si128((__m128i*)(fref2 + (i) * frefstride));
-        T11 = _mm_loadu_si128((__m128i*)(fref2 + (i + 1) * frefstride));
-        T12 = _mm_loadu_si128((__m128i*)(fref2 + (i + 2) * frefstride));
-        T13 = _mm_loadu_si128((__m128i*)(fref2 + (i + 3) * frefstride));
-
-        T14 = _mm_loadu_si128((__m128i*)(fref2 + (i) * frefstride + 16));
-        T15 = _mm_loadu_si128((__m128i*)(fref2 + (i + 1) * frefstride + 16));
-        T16 = _mm_loadu_si128((__m128i*)(fref2 + (i + 2) * frefstride + 16));
-        T17 = _mm_loadu_si128((__m128i*)(fref2 + (i + 3) * frefstride + 16));
-
-        T20 = _mm_sad_epu8(T00, T10);
-        T21 = _mm_sad_epu8(T01, T11);
-        T22 = _mm_sad_epu8(T02, T12);
-        T23 = _mm_sad_epu8(T03, T13);
-
-        T24 = _mm_sad_epu8(T04, T14);
-        T25 = _mm_sad_epu8(T05, T15);
-        T26 = _mm_sad_epu8(T06, T16);
-        T27 = _mm_sad_epu8(T07, T17);
-
-        T20 = _mm_add_epi16(T20, T21);
-        T22 = _mm_add_epi16(T22, T23);
-
-        T24 = _mm_add_epi16(T24, T25);
-        T26 = _mm_add_epi16(T26, T27);
-        
-        sum0 = _mm_add_epi16(T20, T22);
-        sum0 = _mm_add_epi16(sum0, T24);
-        sum0 = _mm_add_epi16(sum0, T26);
-
-        sum1 = _mm_shuffle_epi32(sum0, 2);
-        sum0 = _mm_add_epi32(sum0, sum1);
-        res[1] += _mm_cvtsi128_si32(sum0);
-
-        T10 = _mm_loadu_si128((__m128i*)(fref3 + (i) * frefstride));
-        T11 = _mm_loadu_si128((__m128i*)(fref3 + (i + 1) * frefstride));
-        T12 = _mm_loadu_si128((__m128i*)(fref3 + (i + 2) * frefstride));
-        T13 = _mm_loadu_si128((__m128i*)(fref3 + (i + 3) * frefstride));
-
-        T14 = _mm_loadu_si128((__m128i*)(fref3 + (i) * frefstride + 16));
-        T15 = _mm_loadu_si128((__m128i*)(fref3 + (i + 1) * frefstride + 16));
-        T16 = _mm_loadu_si128((__m128i*)(fref3 + (i + 2) * frefstride + 16));
-        T17 = _mm_loadu_si128((__m128i*)(fref3 + (i + 3) * frefstride + 16));
-
-        T20 = _mm_sad_epu8(T00, T10);
-        T21 = _mm_sad_epu8(T01, T11);
-        T22 = _mm_sad_epu8(T02, T12);
-        T23 = _mm_sad_epu8(T03, T13);
-
-        T24 = _mm_sad_epu8(T04, T14);
-        T25 = _mm_sad_epu8(T05, T15);
-        T26 = _mm_sad_epu8(T06, T16);
-        T27 = _mm_sad_epu8(T07, T17);
-
-        T20 = _mm_add_epi16(T20, T21);
-        T22 = _mm_add_epi16(T22, T23);
-
-        T24 = _mm_add_epi16(T24, T25);
-        T26 = _mm_add_epi16(T26, T27);
-        
-        sum0 = _mm_add_epi16(T20, T22);
-        sum0 = _mm_add_epi16(sum0, T24);
-        sum0 = _mm_add_epi16(sum0, T26);
-
-        sum1 = _mm_shuffle_epi32(sum0, 2);
-        sum0 = _mm_add_epi32(sum0, sum1);
-        res[2] += _mm_cvtsi128_si32(sum0);
-
-        T10 = _mm_loadu_si128((__m128i*)(fref4 + (i) * frefstride));
-        T11 = _mm_loadu_si128((__m128i*)(fref4 + (i + 1) * frefstride));
-        T12 = _mm_loadu_si128((__m128i*)(fref4 + (i + 2) * frefstride));
-        T13 = _mm_loadu_si128((__m128i*)(fref4 + (i + 3) * frefstride));
-
-        T14 = _mm_loadu_si128((__m128i*)(fref4 + (i) * frefstride + 16));
-        T15 = _mm_loadu_si128((__m128i*)(fref4 + (i + 1) * frefstride + 16));
-        T16 = _mm_loadu_si128((__m128i*)(fref4 + (i + 2) * frefstride + 16));
-        T17 = _mm_loadu_si128((__m128i*)(fref4 + (i + 3) * frefstride + 16));
-
-        T20 = _mm_sad_epu8(T00, T10);
-        T21 = _mm_sad_epu8(T01, T11);
-        T22 = _mm_sad_epu8(T02, T12);
-        T23 = _mm_sad_epu8(T03, T13);
-
-        T24 = _mm_sad_epu8(T04, T14);
-        T25 = _mm_sad_epu8(T05, T15);
-        T26 = _mm_sad_epu8(T06, T16);
-        T27 = _mm_sad_epu8(T07, T17);
-
-        T20 = _mm_add_epi16(T20, T21);
-        T22 = _mm_add_epi16(T22, T23);
-
-        T24 = _mm_add_epi16(T24, T25);
-        T26 = _mm_add_epi16(T26, T27);
-        
-        sum0 = _mm_add_epi16(T20, T22);
-        sum0 = _mm_add_epi16(sum0, T24);
-        sum0 = _mm_add_epi16(sum0, T26);
-
-        sum1 = _mm_shuffle_epi32(sum0, 2);
-        sum0 = _mm_add_epi32(sum0, sum1);
-        res[3] += _mm_cvtsi128_si32(sum0);
-
-        T00 = _mm_load_si128((__m128i*)(fenc + (i + 4) * FENC_STRIDE));
-        T01 = _mm_load_si128((__m128i*)(fenc + (i + 5) * FENC_STRIDE));
-        T02 = _mm_load_si128((__m128i*)(fenc + (i + 6) * FENC_STRIDE));
-        T03 = _mm_load_si128((__m128i*)(fenc + (i + 7) * FENC_STRIDE));
-
-        T04 = _mm_load_si128((__m128i*)(fenc + (i + 4) * FENC_STRIDE + 16));
-        T05 = _mm_load_si128((__m128i*)(fenc + (i + 5) * FENC_STRIDE + 16));
-        T06 = _mm_load_si128((__m128i*)(fenc + (i + 6) * FENC_STRIDE + 16));
-        T07 = _mm_load_si128((__m128i*)(fenc + (i + 7) * FENC_STRIDE + 16));
-
-        T10 = _mm_loadu_si128((__m128i*)(fref1 + (i + 4) * frefstride));
-        T11 = _mm_loadu_si128((__m128i*)(fref1 + (i + 5) * frefstride));
-        T12 = _mm_loadu_si128((__m128i*)(fref1 + (i + 6) * frefstride));
-        T13 = _mm_loadu_si128((__m128i*)(fref1 + (i + 7) * frefstride));
-
-        T14 = _mm_loadu_si128((__m128i*)(fref1 + (i + 4) * frefstride + 16));
-        T15 = _mm_loadu_si128((__m128i*)(fref1 + (i + 5) * frefstride + 16));
-        T16 = _mm_loadu_si128((__m128i*)(fref1 + (i + 6) * frefstride + 16));
-        T17 = _mm_loadu_si128((__m128i*)(fref1 + (i + 7) * frefstride + 16));
-
-        T20 = _mm_sad_epu8(T00, T10);
-        T21 = _mm_sad_epu8(T01, T11);
-        T22 = _mm_sad_epu8(T02, T12);
-        T23 = _mm_sad_epu8(T03, T13);
-
-        T24 = _mm_sad_epu8(T04, T14);
-        T25 = _mm_sad_epu8(T05, T15);
-        T26 = _mm_sad_epu8(T06, T16);
-        T27 = _mm_sad_epu8(T07, T17);
-
-        T20 = _mm_add_epi16(T20, T21);
-        T22 = _mm_add_epi16(T22, T23);
-
-        T24 = _mm_add_epi16(T24, T25);
-        T26 = _mm_add_epi16(T26, T27);
-        
-        sum0 = _mm_add_epi16(T20, T22);
-        sum0 = _mm_add_epi16(sum0, T24);
-        sum0 = _mm_add_epi16(sum0, T26);
-
-        sum1 = _mm_shuffle_epi32(sum0, 2);
-        sum0 = _mm_add_epi32(sum0, sum1);
-        res[0] += _mm_cvtsi128_si32(sum0);
-
-        T10 = _mm_loadu_si128((__m128i*)(fref2 + (i + 4) * frefstride));
-        T11 = _mm_loadu_si128((__m128i*)(fref2 + (i + 5) * frefstride));
-        T12 = _mm_loadu_si128((__m128i*)(fref2 + (i + 6) * frefstride));
-        T13 = _mm_loadu_si128((__m128i*)(fref2 + (i + 7) * frefstride));
-
-        T14 = _mm_loadu_si128((__m128i*)(fref2 + (i + 4) * frefstride + 16));
-        T15 = _mm_loadu_si128((__m128i*)(fref2 + (i + 5) * frefstride + 16));
-        T16 = _mm_loadu_si128((__m128i*)(fref2 + (i + 6) * frefstride + 16));
-        T17 = _mm_loadu_si128((__m128i*)(fref2 + (i + 7) * frefstride + 16));
-
-        T20 = _mm_sad_epu8(T00, T10);
-        T21 = _mm_sad_epu8(T01, T11);
-        T22 = _mm_sad_epu8(T02, T12);
-        T23 = _mm_sad_epu8(T03, T13);
-
-        T24 = _mm_sad_epu8(T04, T14);
-        T25 = _mm_sad_epu8(T05, T15);
-        T26 = _mm_sad_epu8(T06, T16);
-        T27 = _mm_sad_epu8(T07, T17);
-
-        T20 = _mm_add_epi16(T20, T21);
-        T22 = _mm_add_epi16(T22, T23);
-
-        T24 = _mm_add_epi16(T24, T25);
-        T26 = _mm_add_epi16(T26, T27);
-        
-        sum0 = _mm_add_epi16(T20, T22);
-        sum0 = _mm_add_epi16(sum0, T24);
-        sum0 = _mm_add_epi16(sum0, T26);
-
-        sum1 = _mm_shuffle_epi32(sum0, 2);
-        sum0 = _mm_add_epi32(sum0, sum1);
-        res[1] += _mm_cvtsi128_si32(sum0);
-
-        T10 = _mm_loadu_si128((__m128i*)(fref3 + (i + 4) * frefstride));
-        T11 = _mm_loadu_si128((__m128i*)(fref3 + (i + 5) * frefstride));
-        T12 = _mm_loadu_si128((__m128i*)(fref3 + (i + 6) * frefstride));
-        T13 = _mm_loadu_si128((__m128i*)(fref3 + (i + 7) * frefstride));
-
-        T14 = _mm_loadu_si128((__m128i*)(fref3 + (i + 4) * frefstride + 16));
-        T15 = _mm_loadu_si128((__m128i*)(fref3 + (i + 5) * frefstride + 16));
-        T16 = _mm_loadu_si128((__m128i*)(fref3 + (i + 6) * frefstride + 16));
-        T17 = _mm_loadu_si128((__m128i*)(fref3 + (i + 7) * frefstride + 16));
-
-        T20 = _mm_sad_epu8(T00, T10);
-        T21 = _mm_sad_epu8(T01, T11);
-        T22 = _mm_sad_epu8(T02, T12);
-        T23 = _mm_sad_epu8(T03, T13);
-
-        T24 = _mm_sad_epu8(T04, T14);
-        T25 = _mm_sad_epu8(T05, T15);
-        T26 = _mm_sad_epu8(T06, T16);
-        T27 = _mm_sad_epu8(T07, T17);
-
-        T20 = _mm_add_epi16(T20, T21);
-        T22 = _mm_add_epi16(T22, T23);
-
-        T24 = _mm_add_epi16(T24, T25);
-        T26 = _mm_add_epi16(T26, T27);
-        
-        sum0 = _mm_add_epi16(T20, T22);
-        sum0 = _mm_add_epi16(sum0, T24);
-        sum0 = _mm_add_epi16(sum0, T26);
-
-        sum1 = _mm_shuffle_epi32(sum0, 2);
-        sum0 = _mm_add_epi32(sum0, sum1);
-        res[2] += _mm_cvtsi128_si32(sum0);
-
-        T10 = _mm_loadu_si128((__m128i*)(fref4 + (i + 4) * frefstride));
-        T11 = _mm_loadu_si128((__m128i*)(fref4 + (i + 5) * frefstride));
-        T12 = _mm_loadu_si128((__m128i*)(fref4 + (i + 6) * frefstride));
-        T13 = _mm_loadu_si128((__m128i*)(fref4 + (i + 7) * frefstride));
-
-        T14 = _mm_loadu_si128((__m128i*)(fref4 + (i + 4) * frefstride + 16));
-        T15 = _mm_loadu_si128((__m128i*)(fref4 + (i + 5) * frefstride + 16));
-        T16 = _mm_loadu_si128((__m128i*)(fref4 + (i + 6) * frefstride + 16));
-        T17 = _mm_loadu_si128((__m128i*)(fref4 + (i + 7) * frefstride + 16));
-
-        T20 = _mm_sad_epu8(T00, T10);
-        T21 = _mm_sad_epu8(T01, T11);
-        T22 = _mm_sad_epu8(T02, T12);
-        T23 = _mm_sad_epu8(T03, T13);
-
-        T24 = _mm_sad_epu8(T04, T14);
-        T25 = _mm_sad_epu8(T05, T15);
-        T26 = _mm_sad_epu8(T06, T16);
-        T27 = _mm_sad_epu8(T07, T17);
-
-        T20 = _mm_add_epi16(T20, T21);
-        T22 = _mm_add_epi16(T22, T23);
-
-        T24 = _mm_add_epi16(T24, T25);
-        T26 = _mm_add_epi16(T26, T27);
-        
-        sum0 = _mm_add_epi16(T20, T22);
-        sum0 = _mm_add_epi16(sum0, T24);
-        sum0 = _mm_add_epi16(sum0, T26);
-
-        sum1 = _mm_shuffle_epi32(sum0, 2);
-        sum0 = _mm_add_epi32(sum0, sum1);
-        res[3] += _mm_cvtsi128_si32(sum0);
+        PROCESS_32x4x4(i);
+        PROCESS_32x4x4(i + 4);
     }
 }
 


More information about the x265-devel mailing list