[x265] [PATCH] asm: avx2 code for sad 16x4, 16x8, 16x12, 16x16, 16x32 for 10bpp

[sumalatha at multicorewareinc.com]

# HG changeset patch
# User Sumalatha Polureddy
# Date 1430807066 -19800
#      Tue May 05 11:54:26 2015 +0530
# Node ID 1cee00d68a2ace6b4da51d4a18a3052bd94987fc
# Parent  f32e6464225afa02983af1b1905f50cdccae5244
asm: avx2 code for sad 16x4,16x8,16x12,16x16,16x32 for 10bpp

sse2:
sad[ 16x4]  2.78x    274.25          761.77
sad[ 16x8]  3.08x    455.14          1401.20
sad[16x12]  3.16x    644.93          2039.52
sad[16x16]  3.28x    830.14          2725.86
sad[16x32]  3.38x    1603.12         5415.46
avx2:
sad[ 16x4]  5.38x    140.91          758.03
sad[ 16x8]  6.14x    214.04          1313.60
sad[16x12]  6.44x    316.88          2039.84
sad[16x16]  6.82x    396.83          2705.34
sad[16x32]  7.13x    738.81          5268.11

diff -r f32e6464225a -r 1cee00d68a2a source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Mon May 04 15:15:42 2015 -0500
+++ b/source/common/x86/asm-primitives.cpp	Tue May 05 11:54:26 2015 +0530
@@ -1239,7 +1239,11 @@
         p.chroma[X265_CSP_I422].cu[BLOCK_422_16x32].sub_ps = x265_pixel_sub_ps_16x32_avx2;
         p.chroma[X265_CSP_I422].cu[BLOCK_422_32x64].sub_ps = x265_pixel_sub_ps_32x64_avx2;
 
+        p.pu[LUMA_16x4].sad = x265_pixel_sad_16x4_avx2;
         p.pu[LUMA_16x8].sad = x265_pixel_sad_16x8_avx2;
+        p.pu[LUMA_16x12].sad = x265_pixel_sad_16x12_avx2;
+        p.pu[LUMA_16x16].sad = x265_pixel_sad_16x16_avx2;
+        p.pu[LUMA_16x32].sad = x265_pixel_sad_16x32_avx2;
 
         p.pu[LUMA_16x4].convert_p2s = x265_filterPixelToShort_16x4_avx2;
         p.pu[LUMA_16x8].convert_p2s = x265_filterPixelToShort_16x8_avx2;
diff -r f32e6464225a -r 1cee00d68a2a source/common/x86/sad16-a.asm
--- a/source/common/x86/sad16-a.asm	Mon May 04 15:15:42 2015 -0500
+++ b/source/common/x86/sad16-a.asm	Tue May 05 11:54:26 2015 +0530
@@ -385,6 +385,13 @@
 SAD  8, 16
 SAD  8, 32
 
+INIT_YMM avx2
+SAD  16,  4
+SAD  16,  8
+SAD  16, 12
+SAD  16, 16
+SAD  16, 32
+
 ;------------------------------------------------------------------
 ; int pixel_sad_32xN( uint16_t *, intptr_t, uint16_t *, intptr_t )
 ;------------------------------------------------------------------
@@ -772,58 +779,6 @@
 %endif
     movd     eax, xm0
     RET
-
-INIT_YMM avx2
-cglobal pixel_sad_16x8, 4,7,7
-    xorps           m0, m0
-    xorps           m6, m6
-    movu            m5, [pw_1]
-    mov             r4d, 8/4
-    add             r1d, r1d
-    add             r3d, r3d
-    lea             r5, [r1 * 3]
-    lea             r6, [r3 * 3]
-.loop
-    movu            m1, [r0]               ; row 0 of pix0
-    movu            m2, [r2]               ; row 0 of pix1
-    movu            m3, [r0 + r1]          ; row 1 of pix0
-    movu            m4, [r2 + r3]          ; row 1 of pix1
-    psubw           m1, m2
-    pabsw           m1, m1
-    pmaddwd         m1, m5
-    psubw           m3, m4
-    pabsw           m3, m3
-    pmaddwd         m3, m5
-    paddd           m0, m1
-    paddd           m6, m3
-
-    movu            m1, [r0 + 2 * r1]      ; row 2 of pix0
-    movu            m2, [r2 + 2 * r3]      ; row 2 of pix1
-    movu            m3, [r0 + r5]          ; row 3 of pix0
-    movu            m4, [r2 + r6]          ; row 3 of pix1
-    psubw           m1, m2
-    pabsw           m1, m1
-    pmaddwd         m1, m5
-    psubw           m3, m4
-    pabsw           m3, m3
-    pmaddwd         m3, m5
-    paddd           m0, m1
-    paddd           m6, m3
-    lea     r2,     [r2 + 4 * r3]
-    lea     r0,     [r0 + 4 * r1]
-
-    dec             r4d
-    jnz             .loop
-
-    paddd           m0, m6
-    vextracti128    xm1, m0, 1
-    paddd           xm0, xm1
-    movhlps         xm1, xm0
-    paddd           xm0, xm1
-    pshufd          xm1, xm0, 1
-    paddd           xm0, xm1
-    movd            eax, xm0
-    RET
 ;-----------------------------------------------------------------------------
 ; void pixel_sad_xN_WxH( uint16_t *fenc, uint16_t *pix0, uint16_t *pix1,
 ;                        uint16_t *pix2, intptr_t i_stride, int scores[3] )