[x265] [PATCH] asm: ssse3 code for chroma_p2s for i420, i422, i444, reuse the luma code

[rajesh at multicorewareinc.com]

# HG changeset patch
# User Rajesh Paulraj<rajesh at multicorewareinc.com>
# Date 1427893006 -19800
#      Wed Apr 01 18:26:46 2015 +0530
# Node ID c42c6f1658cbd8e7408a06b92f17a63255975e5f
# Parent  f3fe51f03aa36223c312f0627d7bbccc6fcdde90
asm: ssse3 code for chroma_p2s for i420, i422, i444, reuse the luma code

diff -r f3fe51f03aa3 -r c42c6f1658cb source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Wed Apr 01 18:09:47 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Wed Apr 01 18:26:46 2015 +0530
@@ -1267,6 +1267,43 @@
         p.pu[LUMA_64x48].convert_p2s = x265_filterPixelToShort_64x48_ssse3;
         p.pu[LUMA_64x64].convert_p2s = x265_filterPixelToShort_64x64_ssse3;
 
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_8x4].chroma_p2s = x265_filterPixelToShort_8x4_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_8x8].chroma_p2s = x265_filterPixelToShort_8x8_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_8x16].chroma_p2s = x265_filterPixelToShort_8x16_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_8x32].chroma_p2s = x265_filterPixelToShort_8x32_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_16x4].chroma_p2s = x265_filterPixelToShort_16x4_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_16x8].chroma_p2s = x265_filterPixelToShort_16x8_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_16x12].chroma_p2s = x265_filterPixelToShort_16x12_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_16x16].chroma_p2s = x265_filterPixelToShort_16x16_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_16x32].chroma_p2s = x265_filterPixelToShort_16x32_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_32x8].chroma_p2s = x265_filterPixelToShort_32x8_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_32x16].chroma_p2s = x265_filterPixelToShort_32x16_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_32x24].chroma_p2s = x265_filterPixelToShort_32x24_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_32x32].chroma_p2s = x265_filterPixelToShort_32x32_ssse3;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_8x4].chroma_p2s = x265_filterPixelToShort_8x4_ssse3;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_8x8].chroma_p2s = x265_filterPixelToShort_8x8_ssse3;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_8x16].chroma_p2s = x265_filterPixelToShort_8x16_ssse3;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_8x32].chroma_p2s = x265_filterPixelToShort_8x32_ssse3;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x8].chroma_p2s = x265_filterPixelToShort_16x8_ssse3;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x16].chroma_p2s = x265_filterPixelToShort_16x16_ssse3;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x32].chroma_p2s = x265_filterPixelToShort_16x32_ssse3;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_16x64].chroma_p2s = x265_filterPixelToShort_16x64_ssse3;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_32x16].chroma_p2s = x265_filterPixelToShort_32x16_ssse3;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_32x32].chroma_p2s = x265_filterPixelToShort_32x32_ssse3;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_32x64].chroma_p2s = x265_filterPixelToShort_32x64_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_8x4].chroma_p2s = x265_filterPixelToShort_8x4_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_8x8].chroma_p2s = x265_filterPixelToShort_8x8_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_8x16].chroma_p2s = x265_filterPixelToShort_8x16_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_8x32].chroma_p2s = x265_filterPixelToShort_8x32_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_16x4].chroma_p2s = x265_filterPixelToShort_16x4_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_16x8].chroma_p2s = x265_filterPixelToShort_16x8_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_16x12].chroma_p2s = x265_filterPixelToShort_16x12_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_16x16].chroma_p2s = x265_filterPixelToShort_16x16_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_16x32].chroma_p2s = x265_filterPixelToShort_16x32_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_32x8].chroma_p2s = x265_filterPixelToShort_32x8_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_32x16].chroma_p2s = x265_filterPixelToShort_32x16_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_32x24].chroma_p2s = x265_filterPixelToShort_32x24_ssse3;
+        p.chroma[X265_CSP_I444].pu[LUMA_32x32].chroma_p2s = x265_filterPixelToShort_32x32_ssse3;
 
         p.dst4x4 = x265_dst4_ssse3;
         p.cu[BLOCK_8x8].idct = x265_idct8_ssse3;
@@ -1320,6 +1357,16 @@
         p.pu[LUMA_4x8].convert_p2s = x265_filterPixelToShort_4x8_sse4;
         p.pu[LUMA_4x16].convert_p2s = x265_filterPixelToShort_4x16_sse4;
 
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_4x4].chroma_p2s = x265_filterPixelToShort_4x4_sse4;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_4x8].chroma_p2s = x265_filterPixelToShort_4x8_sse4;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_4x16].chroma_p2s = x265_filterPixelToShort_4x16_sse4;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_4x4].chroma_p2s = x265_filterPixelToShort_4x4_sse4;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_4x8].chroma_p2s = x265_filterPixelToShort_4x8_sse4;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_4x16].chroma_p2s = x265_filterPixelToShort_4x16_sse4;
+        p.chroma[X265_CSP_I444].pu[LUMA_4x4].chroma_p2s = x265_filterPixelToShort_4x4_sse4;
+        p.chroma[X265_CSP_I444].pu[LUMA_4x8].chroma_p2s = x265_filterPixelToShort_4x8_sse4;
+        p.chroma[X265_CSP_I444].pu[LUMA_4x16].chroma_p2s = x265_filterPixelToShort_4x16_sse4;
+
         LUMA_CU_BLOCKCOPY(ps, sse4);
         CHROMA_420_CU_BLOCKCOPY(ps, sse4);
         CHROMA_422_CU_BLOCKCOPY(ps, sse4);