Add Neon implementation of vpx_sad_skip_<w>x<h>x4d functions

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

Also add corresponding unit tests.

Change-Id: Ieb77661ea2bbe357529862a5fb54956e34e8d758

2 files changed, 45 insertions, 11 deletions

diff --git a/vpx_dsp/arm/sad4d_neon.c b/vpx_dsp/arm/sad4d_neon.c
index 6ad6c9621..44cd99028 100644
--- a/vpx_dsp/arm/sad4d_neon.c
+++ b/vpx_dsp/arm/sad4d_neon.c
@@ -308,3 +308,35 @@ SAD_WXH_4D_NEON(64, 32)
 SAD_WXH_4D_NEON(64, 64)
 
 #undef SAD_WXH_4D_NEON
+
+#define SAD_SKIP_WXH_4D_NEON(w, h)                                          \
+  void vpx_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]) {  \
+    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;                                                           \
+  }
+
+SAD_SKIP_WXH_4D_NEON(4, 4)
+SAD_SKIP_WXH_4D_NEON(4, 8)
+
+SAD_SKIP_WXH_4D_NEON(8, 4)
+SAD_SKIP_WXH_4D_NEON(8, 8)
+SAD_SKIP_WXH_4D_NEON(8, 16)
+
+SAD_SKIP_WXH_4D_NEON(16, 8)
+SAD_SKIP_WXH_4D_NEON(16, 16)
+SAD_SKIP_WXH_4D_NEON(16, 32)
+
+SAD_SKIP_WXH_4D_NEON(32, 16)
+SAD_SKIP_WXH_4D_NEON(32, 32)
+SAD_SKIP_WXH_4D_NEON(32, 64)
+
+SAD_SKIP_WXH_4D_NEON(64, 32)
+SAD_SKIP_WXH_4D_NEON(64, 64)
+
+#undef SAD_SKIP_WXH_4D_NEON
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 7bea73895..4c5fab318 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -968,41 +968,43 @@ add_proto qw/void vpx_sad4x4x4d/, "const uint8_t *src_ptr, int src_stride, const
 specialize qw/vpx_sad4x4x4d neon msa sse2 mmi/;
 
 add_proto qw/void vpx_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_sad_skip_64x64x4d avx2 sse2/;
+specialize qw/vpx_sad_skip_64x64x4d neon avx2 sse2/;
 
 add_proto qw/void vpx_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_sad_skip_64x32x4d avx2 sse2/;
+specialize qw/vpx_sad_skip_64x32x4d neon avx2 sse2/;
 
 add_proto qw/void vpx_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_sad_skip_32x64x4d avx2 sse2/;
+specialize qw/vpx_sad_skip_32x64x4d neon avx2 sse2/;
 
 add_proto qw/void vpx_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_sad_skip_32x32x4d avx2 sse2/;
+specialize qw/vpx_sad_skip_32x32x4d neon avx2 sse2/;
 
 add_proto qw/void vpx_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_sad_skip_32x16x4d avx2 sse2/;
+specialize qw/vpx_sad_skip_32x16x4d neon avx2 sse2/;
 
 add_proto qw/void vpx_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_sad_skip_16x32x4d sse2/;
+specialize qw/vpx_sad_skip_16x32x4d neon sse2/;
 
 add_proto qw/void vpx_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_sad_skip_16x16x4d sse2/;
+specialize qw/vpx_sad_skip_16x16x4d neon sse2/;
 
 add_proto qw/void vpx_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_sad_skip_16x8x4d sse2/;
+specialize qw/vpx_sad_skip_16x8x4d neon sse2/;
 
 add_proto qw/void vpx_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_sad_skip_8x16x4d sse2/;
+specialize qw/vpx_sad_skip_8x16x4d neon sse2/;
 
 add_proto qw/void vpx_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_sad_skip_8x8x4d sse2/;
+specialize qw/vpx_sad_skip_8x8x4d neon sse2/;
 
 add_proto qw/void vpx_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_sad_skip_8x4x4d neon/;
 
 add_proto qw/void vpx_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_sad_skip_4x8x4d sse2/;
+specialize qw/vpx_sad_skip_4x8x4d neon sse2/;
 
 add_proto qw/void vpx_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_sad_skip_4x4x4d neon/;
 
 add_proto qw/uint64_t vpx_sum_squares_2d_i16/, "const int16_t *src, int stride, int size";
 specialize qw/vpx_sum_squares_2d_i16 neon sse2 msa/;