[x265] [PATCH] asm: avx2 10bit code for chroma_hpp[32xN] for i420

[rajesh at multicorewareinc.com]

# HG changeset patch
# User Rajesh Paulraj<rajesh at multicorewareinc.com>
# Date 1432711490 -19800
#      Wed May 27 12:54:50 2015 +0530
# Node ID 3c2a050acbd7ee16f44f19ce170a701774c0e460
# Parent  0de4de0f5e1a020de61bde7b981d8ab1fcbb6b74
asm: avx2 10bit code for chroma_hpp[32xN] for i420

avx2:
chroma_hpp[ 32x8]       6.38x    2789.28         17793.53
chroma_hpp[32x16]       6.55x    5534.30         36258.52
chroma_hpp[32x24]       6.50x    8265.85         53710.54
chroma_hpp[32x32]       6.88x    11277.01        77575.87

sse4:
chroma_hpp[ 32x8]       5.54x    3379.60         18730.04
chroma_hpp[32x16]       5.64x    16906.61        95327.80
chroma_hpp[32x24]       5.82x    15781.41        91801.38
chroma_hpp[32x32]       5.37x    12911.95        69306.38

diff -r 0de4de0f5e1a -r 3c2a050acbd7 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Wed May 27 12:52:06 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Wed May 27 12:54:50 2015 +0530
@@ -1538,6 +1538,10 @@
         p.chroma[X265_CSP_I420].pu[CHROMA_420_16x12].filter_hpp = x265_interp_4tap_horiz_pp_16x12_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_16x16].filter_hpp = x265_interp_4tap_horiz_pp_16x16_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_16x32].filter_hpp = x265_interp_4tap_horiz_pp_16x32_avx2;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_32x8].filter_hpp = x265_interp_4tap_horiz_pp_32x8_avx2;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_32x16].filter_hpp = x265_interp_4tap_horiz_pp_32x16_avx2;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_32x24].filter_hpp = x265_interp_4tap_horiz_pp_32x24_avx2;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_32x32].filter_hpp = x265_interp_4tap_horiz_pp_32x32_avx2;
 
         if (cpuMask & X265_CPU_BMI2)
             p.scanPosLast = x265_scanPosLast_avx2_bmi2;
diff -r 0de4de0f5e1a -r 3c2a050acbd7 source/common/x86/ipfilter16.asm
--- a/source/common/x86/ipfilter16.asm	Wed May 27 12:52:06 2015 +0530
+++ b/source/common/x86/ipfilter16.asm	Wed May 27 12:54:50 2015 +0530
@@ -2592,6 +2592,80 @@
 IPFILTER_CHROMA_avx2_16xN 16
 IPFILTER_CHROMA_avx2_16xN 32
 
+;-------------------------------------------------------------------------------------------------------------
+; void interp_4tap_horiz_pp(pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx
+;-------------------------------------------------------------------------------------------------------------
+INIT_YMM avx2
+%macro IPFILTER_CHROMA_avx2_32xN 1
+%if ARCH_X86_64
+cglobal interp_4tap_horiz_pp_32x%1, 5,6,9
+    add             r1d, r1d
+    add             r3d, r3d
+    sub             r0, 2
+    mov             r4d, r4m
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeff]
+    vpbroadcastq    m0, [r5 + r4 * 8]
+%else
+    vpbroadcastq    m0, [tab_ChromaCoeff + r4 * 8]
+%endif
+    mova            m1, [interp8_hpp_shuf]
+    vpbroadcastd    m2, [pd_32]
+    pxor            m5, m5
+    mova            m6, [idct8_shuf2]
+    mova            m7, [pw_pixel_max]
+
+    mov             r6d, %1
+.loop:
+%assign x 0
+%rep 2
+    vbroadcasti128  m3, [r0 + x]
+    vbroadcasti128  m4, [r0 + 8 + x]
+    pshufb          m3, m1
+    pshufb          m4, m1
+
+    pmaddwd         m3, m0
+    pmaddwd         m4, m0
+    phaddd          m3, m4
+    paddd           m3, m2
+    psrad           m3, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m3, m3
+    vpermq          m3, m3, q2020
+    pshufb          xm3, xm6                     ; m3 = WORD[7 6 5 4 3 2 1 0]
+
+    vbroadcasti128  m4, [r0 + 16 + x]
+    vbroadcasti128  m8, [r0 + 24 + x]
+    pshufb          m4, m1
+    pshufb          m8, m1
+
+    pmaddwd         m4, m0
+    pmaddwd         m8, m0
+    phaddd          m4, m8
+    paddd           m4, m2
+    psrad           m4, 6                       ; m3 = DWORD[7 6 3 2 5 4 1 0]
+
+    packusdw        m4, m4
+    vpermq          m4, m4, q2020
+    pshufb          xm4, xm6                    ; m3 = WORD[7 6 5 4 3 2 1 0]
+    vinserti128     m3, m3, xm4, 1
+    CLIPW           m3, m5, m7
+    movu            [r2 + x], m3
+    %assign x x+32
+    %endrep
+
+    add             r0, r1
+    add             r2, r3
+    dec             r6d
+    jnz             .loop
+    RET
+%endif
+%endmacro
+IPFILTER_CHROMA_avx2_32xN 8
+IPFILTER_CHROMA_avx2_32xN 16
+IPFILTER_CHROMA_avx2_32xN 24
+IPFILTER_CHROMA_avx2_32xN 32
+
 ;-----------------------------------------------------------------------------------------------------------------
 ; void interp_4tap_vert_%3_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
 ;-----------------------------------------------------------------------------------------------------------------