Add Neon implementations of vpx_highbd_sad_skip_<w>x<h>x4d

Add Neon implementations of high bitdepth downsampling SAD4D
functions for all block sizes.

Also add corresponding unit tests.

Change-Id: Ib0c2f852e269cbd6cbb8f4dfb54349654abb0adb

2 files changed, 47 insertions, 11 deletions

diff --git a/vpx_dsp/arm/highbd_sad4d_neon.c b/vpx_dsp/arm/highbd_sad4d_neon.c
index 280d2087f..62c4685a7 100644
--- a/vpx_dsp/arm/highbd_sad4d_neon.c
+++ b/vpx_dsp/arm/highbd_sad4d_neon.c
@@ -236,3 +236,37 @@ HBD_SAD_WXH_4D_NEON(32, 64)
 
 HBD_SAD_WXH_4D_NEON(64, 32)
 HBD_SAD_WXH_4D_NEON(64, 64)
+
+#undef HBD_SAD_WXH_4D_NEON
+
+#define HBD_SAD_SKIP_WXH_4D_NEON(w, h)                                       \
+  void vpx_highbd_sad_skip_##w##x##h##x4d_neon(                              \
+      const uint8_t *src, int src_stride, const uint8_t *const ref[4],       \
+      int ref_stride, uint32_t res[4]) {                                     \
+    highbd_sad##w##xhx4d_neon(src, 2 * src_stride, ref, 2 * ref_stride, res, \
+                              ((h) >> 1));                                   \
+    res[0] <<= 1;                                                            \
+    res[1] <<= 1;                                                            \
+    res[2] <<= 1;                                                            \
+    res[3] <<= 1;                                                            \
+  }
+
+HBD_SAD_SKIP_WXH_4D_NEON(4, 4)
+HBD_SAD_SKIP_WXH_4D_NEON(4, 8)
+
+HBD_SAD_SKIP_WXH_4D_NEON(8, 4)
+HBD_SAD_SKIP_WXH_4D_NEON(8, 8)
+HBD_SAD_SKIP_WXH_4D_NEON(8, 16)
+
+HBD_SAD_SKIP_WXH_4D_NEON(16, 8)
+HBD_SAD_SKIP_WXH_4D_NEON(16, 16)
+HBD_SAD_SKIP_WXH_4D_NEON(16, 32)
+
+HBD_SAD_SKIP_WXH_4D_NEON(32, 16)
+HBD_SAD_SKIP_WXH_4D_NEON(32, 32)
+HBD_SAD_SKIP_WXH_4D_NEON(32, 64)
+
+HBD_SAD_SKIP_WXH_4D_NEON(64, 32)
+HBD_SAD_SKIP_WXH_4D_NEON(64, 64)
+
+#undef HBD_SAD_SKIP_WXH_4D_NEON
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 4c5fab318..bde011529 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -1202,41 +1202,43 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   specialize qw/vpx_highbd_sad4x4x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad_skip_64x64x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad_skip_64x64x4d sse2 avx2/;
+  specialize qw/vpx_highbd_sad_skip_64x64x4d neon sse2 avx2/;
 
   add_proto qw/void vpx_highbd_sad_skip_64x32x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad_skip_64x32x4d sse2 avx2/;
+  specialize qw/vpx_highbd_sad_skip_64x32x4d neon sse2 avx2/;
 
   add_proto qw/void vpx_highbd_sad_skip_32x64x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad_skip_32x64x4d sse2 avx2/;
+  specialize qw/vpx_highbd_sad_skip_32x64x4d neon sse2 avx2/;
 
   add_proto qw/void vpx_highbd_sad_skip_32x32x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad_skip_32x32x4d sse2 avx2/;
+  specialize qw/vpx_highbd_sad_skip_32x32x4d neon sse2 avx2/;
 
   add_proto qw/void vpx_highbd_sad_skip_32x16x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad_skip_32x16x4d sse2 avx2/;
+  specialize qw/vpx_highbd_sad_skip_32x16x4d neon sse2 avx2/;
 
   add_proto qw/void vpx_highbd_sad_skip_16x32x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad_skip_16x32x4d sse2 avx2/;
+  specialize qw/vpx_highbd_sad_skip_16x32x4d neon sse2 avx2/;
 
   add_proto qw/void vpx_highbd_sad_skip_16x16x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad_skip_16x16x4d sse2 avx2/;
+  specialize qw/vpx_highbd_sad_skip_16x16x4d neon sse2 avx2/;
 
   add_proto qw/void vpx_highbd_sad_skip_16x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad_skip_16x8x4d sse2 avx2/;
+  specialize qw/vpx_highbd_sad_skip_16x8x4d neon sse2 avx2/;
 
   add_proto qw/void vpx_highbd_sad_skip_8x16x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad_skip_8x16x4d sse2/;
+  specialize qw/vpx_highbd_sad_skip_8x16x4d neon sse2/;
 
   add_proto qw/void vpx_highbd_sad_skip_8x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad_skip_8x8x4d sse2/;
+  specialize qw/vpx_highbd_sad_skip_8x8x4d neon sse2/;
 
   add_proto qw/void vpx_highbd_sad_skip_8x4x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
+  specialize qw/vpx_highbd_sad_skip_8x4x4d neon/;
 
   add_proto qw/void vpx_highbd_sad_skip_4x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad_skip_4x8x4d sse2/;
+  specialize qw/vpx_highbd_sad_skip_4x8x4d neon sse2/;
 
   add_proto qw/void vpx_highbd_sad_skip_4x4x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
+  specialize qw/vpx_highbd_sad_skip_4x4x4d neon/;
 
   #
   # Structured Similarity (SSIM)