hadamard highbd sse2: use tran_low_t for coeff

BUG=webm:1365

Change-Id: Ica414007d8412ceebfffa9e58e8416226a3fe934

diff --git a/test/hadamard_test.cc b/test/hadamard_test.cc
index 962cddb518fc30b5765d06999ad66507584c0238..bc09e952d2333b4f63025c4ceb43afdb47af3786 100644 (file)
--- a/test/hadamard_test.cc
+++ b/test/hadamard_test.cc
@@ -145,14 +145,14 @@ TEST_P(Hadamard8x8Test, VaryStride) {
 INSTANTIATE_TEST_CASE_P(C, Hadamard8x8Test,
                         ::testing::Values(&vpx_hadamard_8x8_c));
 
-// TODO(jingning): Remove highbitdepth flag when the SIMD functions are
-// in place and turn on the unit test.
-#if !CONFIG_VP9_HIGHBITDEPTH
 #if HAVE_SSE2
 INSTANTIATE_TEST_CASE_P(SSE2, Hadamard8x8Test,
                         ::testing::Values(&vpx_hadamard_8x8_sse2));
 #endif  // HAVE_SSE2
 
+// TODO(jingning): Remove highbitdepth flag when the SIMD functions are
+// in place and turn on the unit test.
+#if !CONFIG_VP9_HIGHBITDEPTH
 #if HAVE_SSSE3 && ARCH_X86_64
 INSTANTIATE_TEST_CASE_P(SSSE3, Hadamard8x8Test,
                         ::testing::Values(&vpx_hadamard_8x8_ssse3));
@@ -212,7 +212,6 @@ TEST_P(Hadamard16x16Test, VaryStride) {
   }
 }
 
-#if !CONFIG_VP9_HIGHBITDEPTH
 INSTANTIATE_TEST_CASE_P(C, Hadamard16x16Test,
                         ::testing::Values(&vpx_hadamard_16x16_c));
 
@@ -221,6 +220,7 @@ INSTANTIATE_TEST_CASE_P(SSE2, Hadamard16x16Test,
                         ::testing::Values(&vpx_hadamard_16x16_sse2));
 #endif  // HAVE_SSE2
 
+#if !CONFIG_VP9_HIGHBITDEPTH
 #if HAVE_NEON
 INSTANTIATE_TEST_CASE_P(NEON, Hadamard16x16Test,
                         ::testing::Values(&vpx_hadamard_16x16_neon));

diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 3cb2011b83e69213603c84a35f70785952f9f545..133f2fbcd0321a2111b70061259853cc156fedb1 100644 (file)
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -888,10 +888,10 @@ if (vpx_config("CONFIG_VP9_ENCODER") eq "yes") {
 
   if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
     add_proto qw/void vpx_hadamard_8x8/, "const int16_t *src_diff, int src_stride, tran_low_t *coeff";
-    specialize qw/vpx_hadamard_8x8/;
+    specialize qw/vpx_hadamard_8x8 sse2/;
 
     add_proto qw/void vpx_hadamard_16x16/, "const int16_t *src_diff, int src_stride, tran_low_t *coeff";
-    specialize qw/vpx_hadamard_16x16/;
+    specialize qw/vpx_hadamard_16x16 sse2/;
 
     add_proto qw/int vpx_satd/, "const tran_low_t *coeff, int length";
     specialize qw/vpx_satd/;

diff --git a/vpx_dsp/x86/avg_intrin_sse2.c b/vpx_dsp/x86/avg_intrin_sse2.c
index b0a104bad066a3c2272fae7eca40e390d8b32053..955d9ceab136b53c98f4280b9951fce1cb69e1a2 100644 (file)
--- a/vpx_dsp/x86/avg_intrin_sse2.c
+++ b/vpx_dsp/x86/avg_intrin_sse2.c
@@ -11,6 +11,8 @@
 #include <emmintrin.h>
 
 #include "./vpx_dsp_rtcd.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_dsp/x86/fdct.h"
 #include "vpx_ports/mem.h"
 
 void vpx_minmax_8x8_sse2(const uint8_t *s, int p, const uint8_t *d, int dp,
@@ -213,7 +215,7 @@ static void hadamard_col8_sse2(__m128i *in, int iter) {
 }
 
 void vpx_hadamard_8x8_sse2(int16_t const *src_diff, int src_stride,
-                           int16_t *coeff) {
+                           tran_low_t *coeff) {
   __m128i src[8];
   src[0] = _mm_load_si128((const __m128i *)src_diff);
   src[1] = _mm_load_si128((const __m128i *)(src_diff += src_stride));
@@ -227,25 +229,25 @@ void vpx_hadamard_8x8_sse2(int16_t const *src_diff, int src_stride,
   hadamard_col8_sse2(src, 0);
   hadamard_col8_sse2(src, 1);
 
-  _mm_store_si128((__m128i *)coeff, src[0]);
+  store_tran_low(src[0], coeff);
   coeff += 8;
-  _mm_store_si128((__m128i *)coeff, src[1]);
+  store_tran_low(src[1], coeff);
   coeff += 8;
-  _mm_store_si128((__m128i *)coeff, src[2]);
+  store_tran_low(src[2], coeff);
   coeff += 8;
-  _mm_store_si128((__m128i *)coeff, src[3]);
+  store_tran_low(src[3], coeff);
   coeff += 8;
-  _mm_store_si128((__m128i *)coeff, src[4]);
+  store_tran_low(src[4], coeff);
   coeff += 8;
-  _mm_store_si128((__m128i *)coeff, src[5]);
+  store_tran_low(src[5], coeff);
   coeff += 8;
-  _mm_store_si128((__m128i *)coeff, src[6]);
+  store_tran_low(src[6], coeff);
   coeff += 8;
-  _mm_store_si128((__m128i *)coeff, src[7]);
+  store_tran_low(src[7], coeff);
 }
 
 void vpx_hadamard_16x16_sse2(int16_t const *src_diff, int src_stride,
-                             int16_t *coeff) {
+                             tran_low_t *coeff) {
   int idx;
   for (idx = 0; idx < 4; ++idx) {
     int16_t const *src_ptr =
@@ -254,10 +256,10 @@ void vpx_hadamard_16x16_sse2(int16_t const *src_diff, int src_stride,
   }
 
   for (idx = 0; idx < 64; idx += 8) {
-    __m128i coeff0 = _mm_load_si128((const __m128i *)coeff);
-    __m128i coeff1 = _mm_load_si128((const __m128i *)(coeff + 64));
-    __m128i coeff2 = _mm_load_si128((const __m128i *)(coeff + 128));
-    __m128i coeff3 = _mm_load_si128((const __m128i *)(coeff + 192));
+    __m128i coeff0 = load_tran_low(coeff);
+    __m128i coeff1 = load_tran_low(coeff + 64);
+    __m128i coeff2 = load_tran_low(coeff + 128);
+    __m128i coeff3 = load_tran_low(coeff + 192);
 
     __m128i b0 = _mm_add_epi16(coeff0, coeff1);
     __m128i b1 = _mm_sub_epi16(coeff0, coeff1);
@@ -271,13 +273,13 @@ void vpx_hadamard_16x16_sse2(int16_t const *src_diff, int src_stride,
 
     coeff0 = _mm_add_epi16(b0, b2);
     coeff1 = _mm_add_epi16(b1, b3);
-    _mm_store_si128((__m128i *)coeff, coeff0);
-    _mm_store_si128((__m128i *)(coeff + 64), coeff1);
+    store_tran_low(coeff0, coeff);
+    store_tran_low(coeff1, coeff + 64);
 
     coeff2 = _mm_sub_epi16(b0, b2);
     coeff3 = _mm_sub_epi16(b1, b3);
-    _mm_store_si128((__m128i *)(coeff + 128), coeff2);
-    _mm_store_si128((__m128i *)(coeff + 192), coeff3);
+    store_tran_low(coeff2, coeff + 128);
+    store_tran_low(coeff3, coeff + 192);
 
     coeff += 8;
   }