HalfFloat_SSE2 port from C algorithm to SSE2

source/row_gcc.cc

 // VERSION 2
 /*
  *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS. All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
@@ skipping 5348 matching lines @@
   : "+r"(src_argb),  // %0
     "+r"(dst_argb),  // %1
     "+r"(width)      // %2
   : "r"(poly)        // %3
   : "memory", "cc",
     "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
   );
 }
 #endif  // HAS_ARGBPOLYNOMIALROW_AVX2
 
+#ifdef HAS_HALFFLOATROW_SSE2
+void HalfFloatRow_SSE2(const uint16* src, uint16* dst, float scale, int width) {
+  float mult = 1.9259299444e-34f * scale;
+  asm volatile (
+    "movd        %3,%%xmm4                     \n"
+    "pshufd      $0x0,%%xmm4,%%xmm4            \n"
+    "pxor        %%xmm5,%%xmm5                 \n"
+
+    // 16 pixel loop.
+    LABELALIGN
+  "1:                                          \n"
+    "movdqu      " MEMACCESS(0) ",%%xmm0       \n"  // 8 shorts
+    "lea         " MEMLEA(0x10,0) ",%0         \n"
+    "movdqa      %%xmm0,%%xmm1                 \n"
+    "punpcklwd   %%xmm5,%%xmm0                 \n"  // 8 ints in xmm0/1
+    "cvtdq2ps    %%xmm0,%%xmm0                 \n"  // 8 floats
+    "punpckhwd   %%xmm5,%%xmm1                 \n"
+    "cvtdq2ps    %%xmm1,%%xmm1                 \n"
+    "mulps       %%xmm4,%%xmm0                 \n"
+    "mulps       %%xmm4,%%xmm1                 \n"
+    "psrld       $0xd,%%xmm0                   \n"
+    "psrld       $0xd,%%xmm1                   \n"
+    "packssdw    %%xmm1,%%xmm0                 \n"
+    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
+    "lea       " MEMLEA(0x10,1) ",%1           \n"
+    "sub         $0x8,%2                       \n"
+    "jg          1b                            \n"
+  : "+r"(src),   // %0
+    "+r"(dst),   // %1
+    "+r"(width)  // %2
+  : "rm"(mult)   // %3
+  : "memory", "cc",
+    "xmm0", "xmm1", "xmm4", "xmm5"
+  );
+}
+#endif  // HAS_HALFFLOATROW_SSE2
+
 #ifdef HAS_HALFFLOATROW_AVX2
 void HalfFloatRow_AVX2(const uint16* src, uint16* dst, float scale, int width) {
   asm volatile (
    "vbroadcastss  %3, %%ymm4                  \n"
 
     // 16 pixel loop.
     LABELALIGN
   "1:                                          \n"
     "vpmovzxwd   " MEMACCESS(0) ",%%ymm0       \n"  // 8 shorts -> 8 ints
     "vpmovzxwd   " MEMACCESS2(0x10,0) ",%%ymm1 \n"  // 8 more
     "lea         " MEMLEA(0x20,0) ",%0         \n"
     "vcvtdq2ps   %%ymm0,%%ymm0                 \n"
     "vcvtdq2ps   %%ymm1,%%ymm1                 \n"
     "vmulps      %%ymm0,%%ymm4,%%ymm0          \n"
     "vmulps      %%ymm1,%%ymm4,%%ymm1          \n"
     "vcvtps2ph   $3, %%ymm0, %%xmm0            \n"
     "vcvtps2ph   $3, %%ymm1, %%xmm1            \n"
     "vmovdqu     %%xmm0," MEMACCESS(1) "       \n"
     "vmovdqu     %%xmm1," MEMACCESS2(0x10,1) " \n"
     "lea         " MEMLEA(0x20,1) ",%1         \n"
     "sub         $0x10,%2                      \n"
     "jg          1b                            \n"
     "vzeroupper                                \n"
   : "+r"(src),   // %0
     "+r"(dst),   // %1
     "+r"(width)  // %2
   : "x"(scale)   // %3
   : "memory", "cc",
-    "xmm0", "xmm4"
+    "xmm0", "xmm1", "xmm4"
   );
 }
 #endif  // HAS_HALFFLOATROW_AVX2
 
 #ifdef HAS_ARGBCOLORTABLEROW_X86
 // Tranform ARGB pixels with color table.
 void ARGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb,
                            int width) {
   uintptr_t pixel_temp;
   asm volatile (
@@ skipping 150 matching lines @@
   );
 }
 #endif  // HAS_ARGBLUMACOLORTABLEROW_SSSE3
 
 #endif  // defined(__x86_64__) || defined(__i386__)
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif