[x265] [PATCH] asm code for blockfil_s, 4x4

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari
# Date 1383828996 -19800
# Node ID f2af7af43dfcb08135a08e755f654314a89efae7
# Parent  d71f86b1c58b4fc9f8a3ffeaaef45c60f8bcc468
asm code for blockfil_s, 4x4

diff -r d71f86b1c58b -r f2af7af43dfc source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Thu Nov 07 18:16:22 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp	Thu Nov 07 18:26:36 2013 +0530
@@ -361,6 +361,8 @@
         p.luma_copy_sp[LUMA_64x32] = x265_blockcopy_sp_64x32_sse2;
         p.luma_copy_sp[LUMA_64x48] = x265_blockcopy_sp_64x48_sse2;
         p.luma_copy_sp[LUMA_64x64] = x265_blockcopy_sp_64x64_sse2;
+
+        p.blockfill_s[BLOCK_4x4] = x265_blockfil_s_4x4_sse2;
 #if X86_64
         p.satd[LUMA_8x32] = x265_pixel_satd_8x32_sse2;
         p.satd[LUMA_16x4] = x265_pixel_satd_16x4_sse2;
diff -r d71f86b1c58b -r f2af7af43dfc source/common/x86/blockcopy8.asm
--- a/source/common/x86/blockcopy8.asm	Thu Nov 07 18:16:22 2013 +0530
+++ b/source/common/x86/blockcopy8.asm	Thu Nov 07 18:26:36 2013 +0530
@@ -1646,3 +1646,22 @@
 BLOCKCOPY_SP_W64_H1 64, 32
 BLOCKCOPY_SP_W64_H1 64, 48
 BLOCKCOPY_SP_W64_H1 64, 64
+
+;-----------------------------------------------------------------------------
+; void blockfil_s_4x4(int16_t *dest, intptr_t destride, int16_t val)
+;-----------------------------------------------------------------------------
+INIT_XMM sse2
+cglobal blockfil_s_4x4, 3, 3, 1, dest, destStride, val
+
+add        r1,            r1
+
+movd       m0,            r2d
+pshuflw    m0,            m0,         0
+
+movh       [r0],          m0
+movh       [r0 + r1],     m0
+movh       [r0 + 2 * r1], m0
+lea        r0,            [r0 + 2 * r1]
+movh       [r0 + r1],     m0
+
+RET
diff -r d71f86b1c58b -r f2af7af43dfc source/common/x86/pixel.h
--- a/source/common/x86/pixel.h	Thu Nov 07 18:16:22 2013 +0530
+++ b/source/common/x86/pixel.h	Thu Nov 07 18:26:36 2013 +0530
@@ -266,6 +266,8 @@
 DECL_ADS(2, avx2)
 DECL_ADS(1, avx2)
 
+void x265_blockfil_s_4x4_sse2(int16_t *dst, intptr_t dstride, int16_t val);
+
 #undef DECL_PIXELS
 #undef DECL_SUF
 #undef DECL_HEVC_SSD