vpermq for avx2 ARGB4444ToARGB, ARGB1555ToARGB and RGB565ToARGB

R=harryjin@google.com
BUG=libyuv:462

Review URL: https://webrtc-codereview.appspot.com/52759004.

include/libyuv/row.h

@@ -186,9 +186,8 @@ extern "C" {
 
 // The following are available require VS2012.  Port to GCC.
 #if !defined(LIBYUV_DISABLE_X86) && defined(VISUALC_HAS_AVX2)
-// Some AVX2 versions disabled. See libyuv bug 462.
-// #define HAS_ARGB1555TOARGBROW_AVX2
-// #define HAS_ARGB4444TOARGBROW_AVX2
+#define HAS_ARGB1555TOARGBROW_AVX2
+#define HAS_ARGB4444TOARGBROW_AVX2
 #define HAS_ARGBTOARGB1555ROW_AVX2
 #define HAS_ARGBTOARGB4444ROW_AVX2
 #define HAS_ARGBTORGB565DITHERROW_AVX2
@@ -204,7 +203,7 @@ extern "C" {
 #define HAS_NV12TORGB565ROW_AVX2
 #define HAS_NV21TOARGBROW_AVX2
 #define HAS_NV21TORGB565ROW_AVX2
-// #define HAS_RGB565TOARGBROW_AVX2
+#define HAS_RGB565TOARGBROW_AVX2
 #endif
 
 // The following are available on all x86 platforms, but

source/row_win.cc

@@ -608,6 +608,8 @@ void ARGB1555ToARGBRow_AVX2(const uint8* src_argb1555, uint8* dst_argb,
     vpmulhuw   ymm0, ymm0, ymm6    // << 6 * (256 + 8)
     vpand      ymm2, ymm2, ymm7
     vpor       ymm0, ymm0, ymm2    // AG
+    vpermq     ymm0, ymm0, 0xd8    // mutate for unpack
+    vpermq     ymm1, ymm1, 0xd8
     vpunpckhbw ymm2, ymm1, ymm0
     vpunpcklbw ymm1, ymm1, ymm0
     vmovdqu    [eax * 2 + edx], ymm1  // store 8 pixels of ARGB
@@ -644,6 +646,8 @@ void ARGB4444ToARGBRow_AVX2(const uint8* src_argb4444, uint8* dst_argb,
     vpsllw     ymm1, ymm0, 4
     vpor       ymm2, ymm2, ymm3
     vpor       ymm0, ymm0, ymm1
+    vpermq     ymm0, ymm0, 0xd8    // mutate for unpack
+    vpermq     ymm2, ymm2, 0xd8
     vpunpckhbw ymm1, ymm0, ymm2
     vpunpcklbw ymm0, ymm0, ymm2
     vmovdqu    [eax * 2 + edx], ymm0  // store 8 pixels of ARGB