[x265] [PATCH] asm: avx2 10bit code for sign primitive(356.91 -> 242.00)

[rajesh at multicorewareinc.com]

# HG changeset patch
# User Rajesh Paulraj<rajesh at multicorewareinc.com>
# Date 1435219457 -19800
#      Thu Jun 25 13:34:17 2015 +0530
# Node ID 430625004ef81ba9e9e398d4cf12a68a1cd4b664
# Parent  a03487d6295cf89b065eff36e5c1ec4ee4253243
asm: avx2 10bit code for sign primitive(356.91 -> 242.00)

avx2:
calSign  9.08x    242.00          2197.71

sse4:
calSign  6.16x    356.91          2197.63

diff -r a03487d6295c -r 430625004ef8 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Thu Jun 25 13:29:58 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Thu Jun 25 13:34:17 2015 +0530
@@ -1496,6 +1496,7 @@
         p.scale1D_128to64 = PFX(scale1D_128to64_avx2);
         p.scale2D_64to32 = PFX(scale2D_64to32_avx2);
         p.weight_pp = PFX(weight_pp_avx2);
+        p.sign = x265_calculateSign_avx2;
 
         p.cu[BLOCK_16x16].calcresidual = PFX(getResidual16_avx2);
         p.cu[BLOCK_32x32].calcresidual = PFX(getResidual32_avx2);
diff -r a03487d6295c -r 430625004ef8 source/common/x86/loopfilter.asm
--- a/source/common/x86/loopfilter.asm	Thu Jun 25 13:29:58 2015 +0530
+++ b/source/common/x86/loopfilter.asm	Thu Jun 25 13:34:17 2015 +0530
@@ -1466,3 +1466,42 @@
     movu            [r0 + r4], m4
     RET
 %endif
+
+;-----------------------------------------------------------------------------
+; void calSign(int8_t *dst, const pixel *src1, const pixel *src2, const int endX)
+;-----------------------------------------------------------------------------
+%if HIGH_BIT_DEPTH
+INIT_YMM avx2
+cglobal calculateSign, 4, 7, 5
+    mova            m0, [pw_1]
+    mov             r4d, r3d
+    shr             r3d, 4
+    add             r3d, 1
+    mov             r5, r0
+    movu            m4, [r0 + r4]
+
+.loop
+    movu            m1, [r1]        ; m2 = pRec[x]
+    movu            m2, [r2]        ; m3 = pTmpU[x]
+
+    pcmpgtw         m3, m1, m2
+    pcmpgtw         m2, m1
+
+    pand            m3, m0
+    por             m3, m2
+    packsswb        m3, m3
+    vpermq          m3, m3, q3220
+    movu            [r0 ], xm3
+
+    add             r0, 16
+    add             r1, 32
+    add             r2, 32
+    dec             r3d
+    jnz             .loop
+
+    mov             r6, r0
+    sub             r6, r5
+    sub             r4, r6
+    movu            [r0 + r4], m4
+    RET
+%endif