HalfFloat_SSE2 use movd from memory

pshufd requires 16 byte aligned memory or a register.
Use movd to a register to avoid a segfault if memory for float
is misaligned

Bug: libyuv:759
Test: 32 bit build of LibYUVPlanarTest.TestHalfFloatPlane_16bit_denormal
Change-Id: I6fdcc4317453af5acd4700f9d46425bb2f4a205b
Reviewed-on: https://chromium-review.googlesource.com/840459Reviewed-by: Miguel Casas <mcasas@chromium.org>
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1684
+Version: 1685
 License: BSD
 License File: LICENSE
 

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1684
+#define LIBYUV_VERSION 1685
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_

source/row_gcc.cc

@@ -5910,7 +5910,8 @@ static float kScaleBias = 1.9259299444e-34f;
 void HalfFloatRow_SSE2(const uint16* src, uint16* dst, float scale, int width) {
   scale *= kScaleBias;
   asm volatile (
-    "pshufd      $0x0,%3,%%xmm4                \n"
+    "movd        %3,%%xmm4                     \n"
+    "pshufd      $0x0,%%xmm4,%%xmm4            \n"
     "pxor        %%xmm5,%%xmm5                 \n"
     "sub         %0,%1                         \n"
 
@@ -5935,11 +5936,7 @@ void HalfFloatRow_SSE2(const uint16* src, uint16* dst, float scale, int width) {
   : "+r"(src),    // %0
     "+r"(dst),    // %1
     "+r"(width)   // %2
-#if defined(__x86_64__)
-  : "x"(scale)   // %3
-#else
   : "m"(scale)   // %3
-#endif
   : "memory", "cc",
     "xmm2", "xmm3", "xmm4", "xmm5"
   );