[x265] [PATCH] asm: chroma_hps[12x32, 24x64 for i422] high bit depth

[aasaipriya at multicorewareinc.com]

# HG changeset patch
# User Aasaipriya Chandran <aasaipriya at multicorewareinc.com>
# Date 1432875879 -19800
#      Fri May 29 10:34:39 2015 +0530
# Node ID 940faba60247d4a5a79f7420a60f6e9deacb3d91
# Parent  7fccb7d165b1c11af9e3d329fad2d818175fc8c4
asm: chroma_hps[12x32, 24x64 for i422] high bit depth

diff -r 7fccb7d165b1 -r 940faba60247 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Fri May 29 10:21:56 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Fri May 29 10:34:39 2015 +0530
@@ -1585,12 +1585,14 @@
         p.chroma[X265_CSP_I422].pu[CHROMA_422_32x32].filter_hps = x265_interp_4tap_horiz_ps_32x32_avx2;
         p.chroma[X265_CSP_I422].pu[CHROMA_422_32x48].filter_hps = x265_interp_4tap_horiz_ps_32x48_avx2;
         p.chroma[X265_CSP_I422].pu[CHROMA_422_32x16].filter_hps = x265_interp_4tap_horiz_ps_32x16_avx2;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_12x32].filter_hps = x265_interp_4tap_horiz_ps_12x32_avx2;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_24x64].filter_hps = x265_interp_4tap_horiz_ps_24x64_avx2;
         p.chroma[X265_CSP_I444].pu[LUMA_64x64].filter_hps = x265_interp_4tap_horiz_ps_64x64_avx2;
         p.chroma[X265_CSP_I444].pu[LUMA_64x48].filter_hps = x265_interp_4tap_horiz_ps_64x48_avx2;
         p.chroma[X265_CSP_I444].pu[LUMA_64x32].filter_hps = x265_interp_4tap_horiz_ps_64x32_avx2;
         p.chroma[X265_CSP_I444].pu[LUMA_64x16].filter_hps = x265_interp_4tap_horiz_ps_64x16_avx2;
         p.chroma[X265_CSP_I444].pu[LUMA_48x64].filter_hps = x265_interp_4tap_horiz_ps_48x64_avx2;
-
+
         p.chroma[X265_CSP_I420].pu[CHROMA_420_6x8].filter_hpp = x265_interp_4tap_horiz_pp_6x8_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x2].filter_hpp = x265_interp_4tap_horiz_pp_8x2_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x4].filter_hpp = x265_interp_4tap_horiz_pp_8x4_avx2;
diff -r 7fccb7d165b1 -r 940faba60247 source/common/x86/ipfilter16.asm
--- a/source/common/x86/ipfilter16.asm	Fri May 29 10:21:56 2015 +0530
+++ b/source/common/x86/ipfilter16.asm	Fri May 29 10:34:39 2015 +0530
@@ -9132,9 +9132,10 @@
 IPFILTER_CHROMA_PS_16xN_AVX2 64
 IPFILTER_CHROMA_PS_16xN_AVX2 24
 
+%macro IPFILTER_CHROMA_PS_24xN_AVX2 1
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_24x32, 4, 7, 6
+cglobal interp_4tap_horiz_ps_24x%1, 4, 7, 6
     add                 r1d, r1d
     add                 r3d, r3d
     mov                 r4d, r4m
@@ -9154,7 +9155,7 @@
 
     sub                 r0, 2
     test                r5d, r5d
-    mov                 r4d, 32
+    mov                 r4d, %1
     jz                  .loop0
     sub                 r0, r1
     add                 r4d, 3
@@ -9208,10 +9209,15 @@
     jnz                 .loop0
     RET
 %endif
-
+%endmacro
+
+IPFILTER_CHROMA_PS_24xN_AVX2 32
+IPFILTER_CHROMA_PS_24xN_AVX2 64
+
+%macro IPFILTER_CHROMA_PS_12xN_AVX2 1
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
-cglobal interp_4tap_horiz_ps_12x16, 4, 7, 6
+cglobal interp_4tap_horiz_ps_12x%1, 4, 7, 6
     add                 r1d, r1d
     add                 r3d, r3d
     mov                 r4d, r4m
@@ -9231,7 +9237,7 @@
 
     sub                 r0, 2
     test                r5d, r5d
-    mov                 r4d, 16
+    mov                 r4d, %1
     jz                  .loop0
     sub                 r0, r1
     add                 r4d, 3
@@ -9268,6 +9274,11 @@
     jnz                 .loop0
     RET
 %endif
+%endmacro
+
+IPFILTER_CHROMA_PS_12xN_AVX2 16
+IPFILTER_CHROMA_PS_12xN_AVX2 32
+
 %macro IPFILTER_CHROMA_PS_32xN_AVX2 1
 INIT_YMM avx2
 %if ARCH_X86_64 == 1