[x265] [PATCH 098 of 307] x86: AVX512 pixel_avg_weight_64xN

[mythreyi at multicorewareinc.com]

# HG changeset patch
# User Vignesh Vijayakumar
# Date 1503908210 -19800
#      Mon Aug 28 13:46:50 2017 +0530
# Node ID 45e4dd746cfd9380dbe2344a5754a6ff6e9feed5
# Parent  bf199a5eca5be148be8a0c91cd9f2e8e0e908059
x86: AVX512 pixel_avg_weight_64xN

Size  |  AVX2 performance | AVX512 performance
----------------------------------------------
64x16 |      41.70x       |      60.98x
64x32 |      36.75x       |      68.91x
64x48 |      37.31x       |      59.07x
64x64 |      37.92x       |      58.85x

diff -r bf199a5eca5b -r 45e4dd746cfd source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Mon Aug 28 11:58:37 2017 +0530
+++ b/source/common/x86/asm-primitives.cpp	Mon Aug 28 13:46:50 2017 +0530
@@ -4159,6 +4159,11 @@
         p.pu[LUMA_64x48].luma_hpp = PFX(interp_8tap_horiz_pp_64x48_avx512);
         p.pu[LUMA_64x64].luma_hpp = PFX(interp_8tap_horiz_pp_64x64_avx512);
 
+        p.pu[LUMA_64x16].pixelavg_pp = PFX(pixel_avg_64x16_avx512);
+        p.pu[LUMA_64x32].pixelavg_pp = PFX(pixel_avg_64x32_avx512);
+        p.pu[LUMA_64x48].pixelavg_pp = PFX(pixel_avg_64x48_avx512);
+        p.pu[LUMA_64x64].pixelavg_pp = PFX(pixel_avg_64x64_avx512);
+
     }
 #endif
 }
diff -r bf199a5eca5b -r 45e4dd746cfd source/common/x86/mc-a.asm
--- a/source/common/x86/mc-a.asm	Mon Aug 28 11:58:37 2017 +0530
+++ b/source/common/x86/mc-a.asm	Mon Aug 28 13:46:50 2017 +0530
@@ -5020,6 +5020,58 @@
     RET
 %endif
 
+;-----------------------------------------------------------------------------
+;pixel_avg_pp avx512 code start
+;-----------------------------------------------------------------------------
+%macro PROCESS_PIXELAVG_64x4_AVX512 0
+    movu        m0,             [r2]
+    movu        m2,             [r2 + r3]
+    movu        m1,             [r4]
+    movu        m3,             [r4 + r5]
+    pavgb       m0,             m1
+    pavgb       m2,             m3
+    movu        [r0],           m0
+    movu        [r0 + r1],      m2
+
+    movu        m0,             [r2 + 2 * r3]
+    movu        m2,             [r2 + r7]
+    movu        m1,             [r4 + 2 * r5]
+    movu        m3,             [r4 + r8]
+    pavgb       m0,             m1
+    pavgb       m2,             m3
+    movu        [r0 + 2 * r1],  m0
+    movu        [r0 + r6],      m2
+%endmacro
+
+;-------------------------------------------------------------------------------------------------------------------------------
+;void pixelavg_pp(pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int)
+;-------------------------------------------------------------------------------------------------------------------------------
+%if ARCH_X86_64 && BIT_DEPTH == 8
+%macro PIXEL_AVG_64xN_AVX512 1
+INIT_ZMM avx512
+cglobal pixel_avg_64x%1, 6, 9, 4
+    lea         r6, [3 * r1]
+    lea         r7, [3 * r3]
+    lea         r8, [3 * r5]
+
+%rep %1/4 - 1
+    PROCESS_PIXELAVG_64x4_AVX512
+    lea         r2, [r2 + r3 * 4]
+    lea         r4, [r4 + r5 * 4]
+    lea         r0, [r0 + r1 * 4]
+%endrep
+    PROCESS_PIXELAVG_64x4_AVX512
+    RET
+%endmacro
+
+PIXEL_AVG_64xN_AVX512 16
+PIXEL_AVG_64xN_AVX512 32
+PIXEL_AVG_64xN_AVX512 48
+PIXEL_AVG_64xN_AVX512 64
+%endif
+;-----------------------------------------------------------------------------
+;pixel_avg_pp avx512 code end
+;-----------------------------------------------------------------------------
 ;=============================================================================
 ; pixel avg2
 ;=============================================================================