[x265] [PATCH] pixel_add_ps_8x4, asm code

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari
# Date 1384955886 -19800
# Node ID 44dae88dacf16829fc17942e9f8812ad79dd2813
# Parent  0cb3981d40388411fa29a971447f7978ecd0cae8
pixel_add_ps_8x4, asm code

diff -r 0cb3981d4038 -r 44dae88dacf1 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Wed Nov 20 19:20:18 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp	Wed Nov 20 19:28:06 2013 +0530
@@ -644,6 +644,7 @@
         p.chroma_add_ps[X265_CSP_I420][CHROMA_4x8] = x265_pixel_add_ps_4x8_sse4;
         p.chroma_add_ps[X265_CSP_I420][CHROMA_4x16] = x265_pixel_add_ps_4x16_sse4;
         p.chroma_add_ps[X265_CSP_I420][CHROMA_8x2] = x265_pixel_add_ps_8x2_sse4;
+        p.chroma_add_ps[X265_CSP_I420][CHROMA_8x4] = x265_pixel_add_ps_8x4_sse4;
     }
     if (cpuMask & X265_CPU_AVX)
     {
diff -r 0cb3981d4038 -r 44dae88dacf1 source/common/x86/pixeladd8.asm
--- a/source/common/x86/pixeladd8.asm	Wed Nov 20 19:20:18 2013 +0530
+++ b/source/common/x86/pixeladd8.asm	Wed Nov 20 19:28:06 2013 +0530
@@ -187,3 +187,49 @@
 movh        [r0 + r1],     m0
 
 RET
+
+;-----------------------------------------------------------------------------
+; void pixel_add_ps_8x4(pixel *dest, intptr_t destride, pixel *src0, int16_t *scr1, intptr_t srcStride0, intptr_t srcStride1)
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal pixel_add_ps_8x4, 6, 6, 2, dest, destride, src0, scr1, srcStride0, srcStride1
+
+add         r5,            r5
+
+pmovzxbw    m0,            [r2]
+movu        m1,            [r3]
+
+paddw       m0,            m1
+packuswb    m0,            m0
+
+movh        [r0],          m0
+
+pmovzxbw    m0,            [r2 + r4]
+movu        m1,            [r3 + r5]
+
+paddw       m0,            m1
+packuswb    m0,            m0
+
+movh        [r0 + r1],     m0
+
+pmovzxbw    m0,            [r2 + 2 * r4]
+movu        m1,            [r3 + 2 * r5]
+
+paddw       m0,            m1
+packuswb    m0,            m0
+
+movh        [r0 + 2 * r1], m0
+
+lea         r0,            [r0 + 2 * r1]
+lea         r2,            [r2 + 2 * r4]
+lea         r3,            [r3 + 2 * r5]
+
+pmovzxbw    m0,            [r2 + r4]
+movu        m1,            [r3 + r5]
+
+paddw       m0,            m1
+packuswb    m0,            m0
+
+movh        [r0 + r1],     m0
+
+RET