For rounding in chroma MC SSSE3, use 16-byte pw_3/4 instead of reading 8 bytes

and then using movlhps to dup it into the higher half of the register.

Originally committed as revision 26086 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/x86/dsputil_mmx.c

@@ -41,7 +41,7 @@ DECLARE_ALIGNED(8,  const uint64_t, ff_wtwo) = 0x0002000200020002ULL;
 DECLARE_ALIGNED(16, const uint64_t, ff_pdw_80000000)[2] =
 {0x8000000080000000ULL, 0x8000000080000000ULL};
 
-DECLARE_ALIGNED(8,  const uint64_t, ff_pw_3  ) = 0x0003000300030003ULL;
+DECLARE_ALIGNED(8,  const xmm_reg,  ff_pw_3  ) = {0x0003000300030003ULL, 0x0003000300030003ULL};
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_4  ) = {0x0004000400040004ULL, 0x0004000400040004ULL};
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_5  ) = {0x0005000500050005ULL, 0x0005000500050005ULL};
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_8  ) = {0x0008000800080008ULL, 0x0008000800080008ULL};

libavcodec/x86/dsputil_mmx.h

@@ -32,7 +32,7 @@ extern const uint64_t ff_wtwo;
 
 extern const uint64_t ff_pdw_80000000[2];
 
-extern const uint64_t ff_pw_3;
+extern const xmm_reg  ff_pw_3;
 extern const xmm_reg  ff_pw_4;
 extern const xmm_reg  ff_pw_5;
 extern const xmm_reg  ff_pw_8;

libavcodec/x86/h264_chromamc.asm

@@ -530,9 +530,8 @@ cglobal %1_%2_chroma_mc8_%3, 6, 7, 8
     add           r4, 8
     sub           r4, r5          ; 255*x+8 = x<<8 | (8-x)
     movd          m7, r4d
-    movq          m6, [rnd_1d_%2]
+    movdqa        m6, [rnd_1d_%2]
     pshuflw       m7, m7, 0
-    movlhps       m6, m6
     movlhps       m7, m7
 
 .next2xrows
@@ -568,9 +567,8 @@ cglobal %1_%2_chroma_mc8_%3, 6, 7, 8
     add           r5, 8
     sub           r5, r4          ; 255*y+8 = y<<8 | (8-y)
     movd          m7, r5d
-    movq          m6, [rnd_1d_%2]
+    movdqa        m6, [rnd_1d_%2]
     pshuflw       m7, m7, 0
-    movlhps       m6, m6
     movlhps       m7, m7
 
 .next2yrows