Add MSA optimized ARGB4444ToI420 and ARGB4444ToARGB functions

include/libyuv/macros_msa.h

Index: include/libyuv/macros_msa.h
diff --git a/include/libyuv/macros_msa.h b/include/libyuv/macros_msa.h
index 8a81e8213afba38737e1d8def52ac7c47b25139f..5b6f9f7a3fbe0827a166edef5cdf970315336ed8 100644
--- a/include/libyuv/macros_msa.h
+++ b/include/libyuv/macros_msa.h
@@ -18,6 +18,8 @@
 #define LD_B(RTYPE, psrc) *((RTYPE*)(psrc))   /* NOLINT */
 #define LD_UB(...) LD_B(v16u8, __VA_ARGS__)
 
+#define LD_H(RTYPE, psrc) *((RTYPE *) (psrc))
+
 #define ST_B(RTYPE, in, pdst) *((RTYPE*)(pdst)) = (in)  /* NOLINT */
 #define ST_UB(...) ST_B(v16u8, __VA_ARGS__)
 
@@ -40,6 +42,18 @@
 }
 #define LD_UB4(...) LD_B4(v16u8, __VA_ARGS__)
 
+/* Description : Load vectors with 8 halfword elements with stride
+   Arguments   : Inputs  - psrc, stride
+                 Outputs - out0, out1
+   Details     : Load 8 halfword elements in 'out0' from (psrc)
+                 Load 8 halfword elements in 'out1' from (psrc + stride)
+*/
+#define LD_H2(RTYPE, psrc, stride, out0, out1) {  \

[fbarchard1, 2016/10/14 21:35:16]

would it more clear in the calling code to do the 2 loads?

[manojkumar.bhosale, 2016/10/19 11:56:27]

Removed this macro

+  out0 = LD_H(RTYPE, (psrc));                     \
+  out1 = LD_H(RTYPE, (psrc) + (stride));          \
+}
+#define LD_UH2(...) LD_H2(v8u16, __VA_ARGS__)
+
 /* Description : Store two vectors with stride each having 16 'byte' sized
                  elements
    Arguments   : Inputs - in0, in1, pdst, stride