Test C vs NEON for ScaleDown2Box_16

TBR=kjellander@chromium.org
BUG=libyuv:718
TEST=LibYUVScaleTest.TestScaleRowDown2Box_16

Change-Id: Ic74d29d6f14983ff26e8af541ef702a0f8bf3f17
Reviewed-on: https://chromium-review.googlesource.com/616189Reviewed-by: Cheng Wang <wangcheng@google.com>

source/scale_neon64.cc

@@ -1002,7 +1002,7 @@ void ScaleRowDown2Box_16_NEON(const uint16* src_ptr,
                               int dst_width) {
   asm volatile(
       // change the stride to row 2 pointer
-      "add        %1, %1, %0                     \n"
+      "add        %1, %0, %1, lsl #1             \n"  // ptr + stide * 2
       "1:                                        \n"
       "ld1        {v0.8h, v1.8h}, [%0], #32      \n"  // load row 1 and post inc
       "ld1        {v2.8h, v3.8h}, [%1], #32      \n"  // load row 2 and post inc
@@ -1033,8 +1033,7 @@ void ScaleRowUp2_16_NEON(const uint16* src_ptr,
                          uint16* dst,
                          int dst_width) {
   asm volatile(
-      // change the stride to row 2 pointer
-      "add        %1, %1, %0                     \n"
+      "add        %1, %0, %1, lsl #1             \n"  // ptr + stide * 2
       "movi       v20.4h, #1                     \n"
       "movi       v21.4h, #3                     \n"  // constants
       "movi       v22.4h, #9                     \n"

unit_test/scale_test.cc

@@ -457,7 +457,7 @@ extern "C" void ScaleRowUp2_16_NEON(const uint16* src_ptr,
                                     ptrdiff_t src_stride,
                                     uint16* dst,
                                     int dst_width);
-           
+
 TEST_F(LibYUVScaleTest, TestScaleRowUp2_16) {
   SIMD_ALIGNED(uint16 orig_pixels[640 * 2 + 1]);  // 2 rows + 1 pixel overrun
   SIMD_ALIGNED(uint16 dst_pixels_opt[1280]);
@@ -471,13 +471,13 @@ TEST_F(LibYUVScaleTest, TestScaleRowUp2_16) {
     orig_pixels[i] = i;
   }
   ScaleRowUp2_16_NEON(&orig_pixels[0],
-                      640 * 2,
+                      640,
                       &dst_pixels_c[0],
                       1280);
 
   for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
     ScaleRowUp2_16_NEON(&orig_pixels[0],
-                        640 * 2,
+                        640,
                         &dst_pixels_opt[0],
                         1280);
   }
@@ -488,42 +488,57 @@ TEST_F(LibYUVScaleTest, TestScaleRowUp2_16) {
   EXPECT_EQ(dst_pixels_c[0], (0 * 9 + 1 * 3 + 640 * 3 + 641 * 1 + 8) / 16);
   EXPECT_EQ(dst_pixels_c[1279], 800);
 }
+#endif
 
 extern "C" void ScaleRowDown2Box_16_NEON(const uint16* src_ptr,
                                          ptrdiff_t src_stride,
                                          uint16* dst,
                                          int dst_width);
-           
+
 TEST_F(LibYUVScaleTest, TestScaleRowDown2Box_16) {
   SIMD_ALIGNED(uint16 orig_pixels[2560 * 2]);
-  SIMD_ALIGNED(uint16 dst_pixels_opt[1280]);
   SIMD_ALIGNED(uint16 dst_pixels_c[1280]);
+  SIMD_ALIGNED(uint16 dst_pixels_opt[1280]);
 
   memset(orig_pixels, 0, sizeof(orig_pixels));
-  memset(dst_pixels_opt, 1, sizeof(dst_pixels_opt));
-  memset(dst_pixels_c, 2, sizeof(dst_pixels_c));
+  memset(dst_pixels_c, 1, sizeof(dst_pixels_c));
+  memset(dst_pixels_opt, 2, sizeof(dst_pixels_opt));
 
   for (int i = 0; i < 2560 * 2; ++i) {
     orig_pixels[i] = i;
   }
-  ScaleRowDown2Box_16_NEON(&orig_pixels[0],
-                           2560 * 2,
+  ScaleRowDown2Box_16_C(&orig_pixels[0],
+                           2560,
                            &dst_pixels_c[0],
                            1280);
-
   for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
-    ScaleRowDown2Box_16_NEON(&orig_pixels[0],
-                             2560 * 2,
-                             &dst_pixels_opt[0],
-                             1280);
+#if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
+    int has_neon = TestCpuFlag(kCpuHasNEON);
+    if (has_neon) {
+      ScaleRowDown2Box_16_NEON(&orig_pixels[0],
+                               2560,
+                               &dst_pixels_opt[0],
+                               1280);
+    } else {
+      ScaleRowDown2Box_16_C(&orig_pixels[0],
+                            2560,
+                            &dst_pixels_opt[0],
+                            1280);
+    }
+#else
+    ScaleRowDown2Box_16_C(&orig_pixels[0],
+                          2560,
+                          &dst_pixels_opt[0],
+                          1280);
+#endif
   }
 
   for (int i = 0; i < 1280; ++i) {
     EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
   }
-  EXPECT_EQ(dst_pixels_c[0], 1281);
+
+  EXPECT_EQ(dst_pixels_c[0], (0 + 1 + 2560 + 2561 + 2) / 4);
   EXPECT_EQ(dst_pixels_c[1279], 3839);
 }
-#endif  // __aarch64__
 
 }  // namespace libyuv