[x265] [PATCH] pixel8.inc: sad_x3_4 further optimization

praveen at multicorewareinc.com praveen at multicorewareinc.com
Mon Aug 26 12:18:13 CEST 2013


# HG changeset patch
# User praveentiwari
# Date 1377512284 -19800
# Node ID d25b88f5e12802b1f6a4ca74a965fd086088bd44
# Parent  264892c1305c5efc884c376a6512b968728249c4
pixel8.inc: sad_x3_4 further optimization

diff -r 264892c1305c -r d25b88f5e128 source/common/vec/pixel8.inc
--- a/source/common/vec/pixel8.inc	Mon Aug 26 14:43:19 2013 +0530
+++ b/source/common/vec/pixel8.inc	Mon Aug 26 15:48:04 2013 +0530
@@ -1335,192 +1335,330 @@
     __m64 T10, T11, T12, T13, T14, T15, T16, T17;
     __m64 T20, T21, T22, T23, T24, T25, T26, T27;
 
-    if ((ly % 16) == 0)
-    {
-        for (int i = 0; i < ly; i += 16)
-        {
-            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));
-            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));
-            T0 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 8) * FENC_STRIDE));
-            T1 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 9) * FENC_STRIDE));
-            T2 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 10) * FENC_STRIDE));
-            T3 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 11) * FENC_STRIDE));
-            T4 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 12) * FENC_STRIDE));
-            T5 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 13) * FENC_STRIDE));
-            T6 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 14) * FENC_STRIDE));
-            T7 = _mm_cvtsi32_si64(*(int*)(fenc + (i + 15) * 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));
-
-            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 + (i + 8) * frefstride));
-            T11 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 9) * frefstride));
-            T12 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 10) * frefstride));
-            T13 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 11) * frefstride));
-            T14 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 12) * frefstride));
-            T15 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 13) * frefstride));
-            T16 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 14) * frefstride));
-            T17 = _mm_cvtsi32_si64(*(int*)(fref1 + (i + 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);
-            res[0] = _m_to_int(sum0);
-
-            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));
-
-            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 + (i + 8) * frefstride));
-            T11 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 9) * frefstride));
-            T12 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 10) * frefstride));
-            T13 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 11) * frefstride));
-            T14 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 12) * frefstride));
-            T15 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 13) * frefstride));
-            T16 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 14) * frefstride));
-            T17 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 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);
-            res[1] = _m_to_int(sum1);
-
-            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));
-
-            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 + (i + 8) * frefstride));
-            T11 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 9) * frefstride));
-            T12 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 10) * frefstride));
-            T13 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 11) * frefstride));
-            T14 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 12) * frefstride));
-            T15 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 13) * frefstride));
-            T16 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 14) * frefstride));
-            T17 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 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);
-            res[2] = _m_to_int(sum2);
-        }
+    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));
+        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));
+        T12 = _mm_cvtsi32_si64(*(int*)(fref1 + 2 * frefstride));
+        T13 = _mm_cvtsi32_si64(*(int*)(fref1 + 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 + 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));
+
+        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));
+        T12 = _mm_cvtsi32_si64(*(int*)(fref3 + 2 * frefstride));
+        T13 = _mm_cvtsi32_si64(*(int*)(fref3 + 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);
+    }
+    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);
+    }
+    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);
     }
     else if ((ly % 8) == 0)
     {
@@ -1561,7 +1699,6 @@
             sum0 = _mm_add_pi16(sum0, T25);
             sum0 = _mm_add_pi16(sum0, T26);
             sum0 = _mm_add_pi16(sum0, T27);
-            res[0] = _m_to_int(sum0);
 
             T10 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 0) * frefstride));
             T11 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 1) * frefstride));
@@ -1589,7 +1726,6 @@
             sum1 = _mm_add_pi16(sum1, T25);
             sum1 = _mm_add_pi16(sum1, T26);
             sum1 = _mm_add_pi16(sum1, T27);
-            res[1] = _m_to_int(sum1);
 
             T10 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 0) * frefstride));
             T11 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 1) * frefstride));
@@ -1617,7 +1753,6 @@
             sum2 = _mm_add_pi16(sum2, T25);
             sum2 = _mm_add_pi16(sum2, T26);
             sum2 = _mm_add_pi16(sum2, T27);
-            res[2] = _m_to_int(sum2);
         }
     }
     else
@@ -1643,7 +1778,6 @@
             sum0 = _mm_add_pi16(sum0, T21);
             sum0 = _mm_add_pi16(sum0, T22);
             sum0 = _mm_add_pi16(sum0, T23);
-            res[0] = _m_to_int(sum0);
 
             T10 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 0) * frefstride));
             T11 = _mm_cvtsi32_si64(*(int*)(fref2 + (i + 1) * frefstride));
@@ -1659,7 +1793,6 @@
             sum1 = _mm_add_pi16(sum1, T21);
             sum1 = _mm_add_pi16(sum1, T22);
             sum1 = _mm_add_pi16(sum1, T23);
-            res[1] = _m_to_int(sum1);
 
             T10 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 0) * frefstride));
             T11 = _mm_cvtsi32_si64(*(int*)(fref3 + (i + 1) * frefstride));
@@ -1675,9 +1808,11 @@
             sum2 = _mm_add_pi16(sum2, T21);
             sum2 = _mm_add_pi16(sum2, T22);
             sum2 = _mm_add_pi16(sum2, T23);
-            res[2] = _m_to_int(sum2);
         }
     }
+    res[0] = _m_to_int(sum0);
+    res[1] = _m_to_int(sum1);
+    res[2] = _m_to_int(sum2);
 }
 
 #else /* if HAVE_MMX */


More information about the x265-devel mailing list