[x265] [PATCH] pixel_add_ps_6x8, asm code

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari
# Date 1385038314 -19800
# Node ID a02029796cb1c99e75af53844b6b18b74fa68d27
# Parent  907b99bb4a404f4d5dd1655751b47c226b574f53
pixel_add_ps_6x8, asm code

diff -r 907b99bb4a40 -r a02029796cb1 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Thu Nov 21 17:27:06 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp	Thu Nov 21 18:21:54 2013 +0530
@@ -639,6 +639,7 @@
         p.chroma[X265_CSP_I420].add_ps[CHROMA_4x4] = x265_pixel_add_ps_4x4_sse4;
         p.chroma[X265_CSP_I420].add_ps[CHROMA_4x8] = x265_pixel_add_ps_4x8_sse4;
         p.chroma[X265_CSP_I420].add_ps[CHROMA_4x16] = x265_pixel_add_ps_4x16_sse4;
+        p.chroma[X265_CSP_I420].add_ps[CHROMA_6x8] = x265_pixel_add_ps_6x8_sse4;
         p.chroma[X265_CSP_I420].add_ps[CHROMA_8x2] = x265_pixel_add_ps_8x2_sse4;
         p.chroma[X265_CSP_I420].add_ps[CHROMA_8x4] = x265_pixel_add_ps_8x4_sse4;
         p.chroma[X265_CSP_I420].add_ps[CHROMA_8x6] = x265_pixel_add_ps_8x6_sse4;
diff -r 907b99bb4a40 -r a02029796cb1 source/common/x86/pixeladd8.asm
--- a/source/common/x86/pixeladd8.asm	Thu Nov 21 17:27:06 2013 +0530
+++ b/source/common/x86/pixeladd8.asm	Thu Nov 21 18:21:54 2013 +0530
@@ -161,6 +161,71 @@
 PIXEL_ADD_PS_W4_H4   4,  8
 PIXEL_ADD_PS_W4_H4   4, 16
 
+
+;-----------------------------------------------------------------------------
+; void pixel_add_ps_%1x%2(pixel *dest, intptr_t destride, pixel *src0, int16_t *scr1, intptr_t srcStride0, intptr_t srcStride1)
+;-----------------------------------------------------------------------------
+%macro PIXEL_ADD_PS_W6_H4 2
+INIT_XMM sse4
+cglobal pixel_add_ps_%1x%2, 6, 7, 2, dest, destride, src0, scr1, srcStride0, srcStride1
+
+add         r5,            r5
+
+mov         r6d,           %2/4
+
+.loop
+      pmovzxbw    m0,                [r2]
+      movu        m1,                [r3]
+
+      paddw       m0,                m1
+      packuswb    m0,                m0
+
+      movd        [r0],              m0
+      pextrw      [r0 + 4],          m0,        2
+
+      pmovzxbw    m0,                [r2 + r4]
+      movu        m1,                [r3 + r5]
+
+      paddw       m0,                m1
+      packuswb    m0,                m0
+
+      movd        [r0 + r1],         m0
+      pextrw      [r0 + r1 + 4],     m0,        2
+
+      pmovzxbw    m0,                [r2 + 2 * r4]
+      movu        m1,                [r3 + 2 * r5]
+
+      paddw       m0,                m1
+      packuswb    m0,                m0
+
+      movd        [r0 + 2 * r1],     m0
+      pextrw      [r0 + 2 * r1 + 4], m0,        2
+
+      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
+
+      movd        [r0 + r1],         m0
+      pextrw      [r0 + r1 + 4],     m0,        2
+
+      lea         r0,                [r0 + 2 * r1]
+      lea         r2,                [r2 + 2 * r4]
+      lea         r3,                [r3 + 2 * r5]
+
+      dec         r6d
+      jnz         .loop
+
+RET
+%endmacro
+
+PIXEL_ADD_PS_W6_H4 6,  8
+
 ;-----------------------------------------------------------------------------
 ; void pixel_add_ps_8x2(pixel *dest, intptr_t destride, pixel *src0, int16_t *scr1, intptr_t srcStride0, intptr_t srcStride1)
 ;-----------------------------------------------------------------------------