satd highbd neon: use tran_low_t for coeff

BUG=webm:1365

Change-Id: I43521ad32b6c96737a8ef2b8c327f901fd7eaf84

diff --git a/test/avg_test.cc b/test/avg_test.cc
index 09cefe816c45e686d5894cc44780ffbfbf91166d..68a83d66e271c68a45c203d78a8742cef3e81d0d 100644 (file)
--- a/test/avg_test.cc
+++ b/test/avg_test.cc
@@ -379,15 +379,11 @@ INSTANTIATE_TEST_CASE_P(
                       make_tuple(64, &vpx_int_pro_col_neon,
                                  &vpx_int_pro_col_c)));
 
-// TODO(jingning): Remove the highbitdepth flag once the SIMD functions are
-// in place.
-#if !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(NEON, SatdTest,
                         ::testing::Values(make_tuple(16, &vpx_satd_neon),
                                           make_tuple(64, &vpx_satd_neon),
                                           make_tuple(256, &vpx_satd_neon),
                                           make_tuple(1024, &vpx_satd_neon)));
-#endif  // !CONFIG_VP9_HIGHBITDEPTH
 #endif  // HAVE_NEON
 
 #if HAVE_MSA
@@ -414,6 +410,8 @@ INSTANTIATE_TEST_CASE_P(
                       make_tuple(64, &vpx_int_pro_col_msa,
                                  &vpx_int_pro_col_c)));
 
+// TODO(jingning): Remove the highbitdepth flag once the SIMD functions are
+// in place.
 #if !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(MSA, SatdTest,
                         ::testing::Values(make_tuple(16, &vpx_satd_msa),

diff --git a/vpx_dsp/arm/avg_neon.c b/vpx_dsp/arm/avg_neon.c
index 001517d33ee71a8095b515fedb28f453d8971368..cca9a9324230a81c702cb60f6199f34be7250991 100644 (file)
--- a/vpx_dsp/arm/avg_neon.c
+++ b/vpx_dsp/arm/avg_neon.c
@@ -15,6 +15,7 @@
 #include "./vpx_config.h"
 
 #include "vpx/vpx_integer.h"
+#include "vpx_dsp/arm/idct_neon.h"
 
 static INLINE unsigned int horizontal_add_u16x8(const uint16x8_t v_16x8) {
   const uint32x4_t a = vpaddlq_u16(v_16x8);
@@ -64,13 +65,13 @@ unsigned int vpx_avg_8x8_neon(const uint8_t *s, int p) {
 
 // coeff: 16 bits, dynamic range [-32640, 32640].
 // length: value range {16, 64, 256, 1024}.
-int vpx_satd_neon(const int16_t *coeff, int length) {
+int vpx_satd_neon(const tran_low_t *coeff, int length) {
   const int16x4_t zero = vdup_n_s16(0);
   int32x4_t accum = vdupq_n_s32(0);
 
   do {
-    const int16x8_t src0 = vld1q_s16(coeff);
-    const int16x8_t src8 = vld1q_s16(coeff + 8);
+    const int16x8_t src0 = load_tran_low_to_s16q(coeff);
+    const int16x8_t src8 = load_tran_low_to_s16q(coeff + 8);
     accum = vabal_s16(accum, vget_low_s16(src0), zero);
     accum = vabal_s16(accum, vget_high_s16(src0), zero);
     accum = vabal_s16(accum, vget_low_s16(src8), zero);

diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 347e341cd85786004b86d1241ed472be92c87137..95a5c39fe2085cb68184061b259ba755f10e534b 100644 (file)
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -894,7 +894,7 @@ if (vpx_config("CONFIG_VP9_ENCODER") eq "yes") {
     specialize qw/vpx_hadamard_16x16/;
 
     add_proto qw/int vpx_satd/, "const tran_low_t *coeff, int length";
-    specialize qw/vpx_satd sse2/;
+    specialize qw/vpx_satd sse2 neon/;
   } else {
     add_proto qw/void vpx_hadamard_8x8/, "const int16_t *src_diff, int src_stride, int16_t *coeff";
     specialize qw/vpx_hadamard_8x8 sse2 neon msa/, "$ssse3_x86_64";