Convolution horizontal filter SSSE3 optimization

- Apply signal direction/4-pixel vertical/8-pixel vertical
  parallelism.
- Add unit test to verify the bit exact result.
- Overall encoding time improves ~24% on Xeon E5-2680 CPU.

Change-Id: I104dcbfd43451476fee1f94cd16ca5f965878e59

diff --git a/test/test.mk b/test/test.mk
index 9a87a809d9779f6db49319c71b268e2cb7787008..47bc5d5db2de345a75e9f8b45da54f28badc749b 100644 (file)
--- a/test/test.mk
+++ b/test/test.mk
@@ -207,6 +207,7 @@ LIBVPX_TEST_SRCS-$(CONFIG_VP10) += vp10_inv_txfm1d_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP10) += vp10_fwd_txfm2d_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP10) += vp10_inv_txfm2d_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP10) += vp10_convolve_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP10) += vp10_convolve_optimz_test.cc
 
 TEST_INTRA_PRED_SPEED_SRCS-yes := test_intra_pred_speed.cc
 TEST_INTRA_PRED_SPEED_SRCS-yes += ../md5_utils.h ../md5_utils.c

diff --git a/test/vp10_convolve_optimz_test.cc b/test/vp10_convolve_optimz_test.cc
new file mode 100644 (file)
index 0000000..fa460bf
--- /dev/null
+++ b/test/vp10_convolve_optimz_test.cc
@@ -0,0 +1,215 @@
+/*
+ *  Copyright (c) 2016 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vp10_rtcd.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+namespace {
+
+using std::tr1::tuple;
+using libvpx_test::ACMRandom;
+
+typedef void (*conv_horiz_t)(const uint8_t*, int, uint8_t*, int,
+                             int, int, const InterpFilterParams,
+                             const int, int, int);
+// Test parameter list:
+//  <convolve_horiz_func, <width, height>, filter_params, subpel_x_q4, avg>
+typedef tuple<int, int> BlockDimension;
+typedef tuple<conv_horiz_t, BlockDimension, INTERP_FILTER, int, int> ConvParams;
+
+// Note:
+//  src_ and src_ref_ have special boundary requirement
+//  dst_ and dst_ref_ don't
+const size_t maxWidth = 256;
+const size_t maxHeight = 256;
+const size_t maxBlockSize = maxWidth * maxHeight;
+const int horizOffset = 32;
+const int vertiOffset = 32;
+const int stride = 128;
+const int x_step_q4 = 16;
+
+class VP10ConvolveOptimzTest : public ::testing::TestWithParam<ConvParams> {
+ public:
+  virtual ~VP10ConvolveOptimzTest() {}
+  virtual void SetUp() {
+    conv_ = GET_PARAM(0);
+    BlockDimension block = GET_PARAM(1);
+    width_ = std::tr1::get<0>(block);
+    height_ = std::tr1::get<1>(block);
+    filter_ = GET_PARAM(2);
+    subpel_ = GET_PARAM(3);
+    avg_ = GET_PARAM(4);
+
+    alloc_ = new uint8_t[maxBlockSize * 4];
+    src_ = alloc_ + (vertiOffset * maxWidth);
+    src_ += horizOffset;
+    src_ref_ = src_ + maxBlockSize;
+
+    dst_ = alloc_ + 2 * maxBlockSize;
+    dst_ref_ = alloc_ + 3 * maxBlockSize;
+  }
+
+  virtual void TearDown() {
+    delete[] alloc_;
+    libvpx_test::ClearSystemState();
+  }
+
+ protected:
+  void RunHorizFilterBitExactCheck();
+
+ private:
+  void PrepFilterBuffer(uint8_t *src, uint8_t *src_ref,
+                        uint8_t *dst, uint8_t *dst_ref,
+                        int w, int h);
+  void DiffFilterBuffer(const uint8_t *buf, const uint8_t *buf_ref,
+                        int w, int h, int fgroup, int findex);
+  conv_horiz_t conv_;
+  uint8_t *alloc_;
+  uint8_t *src_;
+  uint8_t *dst_;
+  uint8_t *src_ref_;
+  uint8_t *dst_ref_;
+  int width_;
+  int height_;
+  int filter_;
+  int subpel_;
+  int avg_;
+};
+
+void VP10ConvolveOptimzTest::PrepFilterBuffer(uint8_t *src, uint8_t *src_ref,
+                                              uint8_t *dst, uint8_t *dst_ref,
+                                              int w, int h) {
+  int r, c;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+
+  memset(src, 0, maxBlockSize);
+  memset(src_ref, 0, maxBlockSize);
+  memset(dst, 0, maxBlockSize);
+  memset(dst_ref, 0, maxBlockSize);
+
+  uint8_t *src_ptr = src;
+  uint8_t *dst_ptr = dst;
+  uint8_t *src_ref_ptr = src_ref;
+  uint8_t *dst_ref_ptr = dst_ref;
+
+  for (r = 0; r < height_; ++r) {
+    for (c = 0; c < width_; ++c) {
+      src_ptr[c] = rnd.Rand8();
+      src_ref_ptr[c] = src_ptr[c];
+      dst_ptr[c] = rnd.Rand8();
+      dst_ref_ptr[c] = dst_ptr[c];
+    }
+    src_ptr += stride;
+    src_ref_ptr += stride;
+    dst_ptr += stride;
+    dst_ref_ptr += stride;
+  }
+}
+
+void VP10ConvolveOptimzTest::DiffFilterBuffer(const uint8_t *buf,
+                                              const uint8_t *buf_ref,
+                                              int w, int h,
+                                              int filter_group,
+                                              int filter_index) {
+  int r, c;
+  const uint8_t *dst_ptr = buf;
+  const uint8_t *dst_ref_ptr = buf_ref;
+  for (r = 0; r < h; ++r) {
+    for (c = 0; c < w; ++c) {
+      EXPECT_EQ((uint8_t)dst_ref_ptr[c], (uint8_t)dst_ptr[c])
+      << "Error at row: " << r << " col: " << c << " "
+      << "w = " << w << " " << "h = " << h << " "
+      << "filter group index = " << filter_group << " "
+      << "filter index = " << filter_index;
+    }
+    dst_ptr += stride;
+    dst_ref_ptr += stride;
+  }
+}
+
+void VP10ConvolveOptimzTest::RunHorizFilterBitExactCheck() {
+  PrepFilterBuffer(src_, src_ref_, dst_, dst_ref_, width_, height_);
+
+  InterpFilterParams filter_params = vp10_get_interp_filter_params(filter_);
+
+  vp10_convolve_horiz_c(src_ref_, stride, dst_ref_, stride, width_, height_,
+                        filter_params, subpel_, x_step_q4, avg_);
+
+  conv_(src_, stride, dst_, stride, width_, height_,
+        filter_params, subpel_, x_step_q4, avg_);
+
+  DiffFilterBuffer(dst_, dst_ref_, width_, height_, filter_, subpel_);
+
+  // Note:
+  // Here we need calculate a height which is different from the specified one
+  // and test again.
+  int intermediate_height =
+      (((height_ - 1) * 16 + subpel_) >> SUBPEL_BITS) + filter_params.taps;
+  PrepFilterBuffer(src_, src_ref_, dst_, dst_ref_, width_, height_);
+
+  vp10_convolve_horiz_c(src_ref_, stride, dst_ref_, stride, width_,
+                        intermediate_height, filter_params, subpel_, x_step_q4,
+                        avg_);
+
+  conv_(src_, stride, dst_, stride, width_,
+        intermediate_height, filter_params, subpel_, x_step_q4,
+        avg_);
+
+  DiffFilterBuffer(dst_, dst_ref_, width_, intermediate_height, filter_,
+                   subpel_);
+}
+
+TEST_P(VP10ConvolveOptimzTest, HorizBitExactCheck) {
+  RunHorizFilterBitExactCheck();
+}
+
+using std::tr1::make_tuple;
+
+const BlockDimension kBlockDim[] = {
+  make_tuple(4, 4),
+  make_tuple(4, 8),
+  make_tuple(8, 4),
+  make_tuple(8, 8),
+  make_tuple(8, 16),
+  make_tuple(16, 8),
+  make_tuple(16, 16),
+  make_tuple(16, 32),
+  make_tuple(32, 16),
+  make_tuple(32, 32),
+  make_tuple(32, 64),
+  make_tuple(64, 32),
+  make_tuple(64, 64),
+  make_tuple(64, 128),
+  make_tuple(128, 64),
+  make_tuple(128, 128),
+};
+// 10/12-tap filters
+const INTERP_FILTER kFilter[] = {6, 4, 2};
+
+const int kSubpelXQ4[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15};
+
+const int kAvg[] = {0, 1};
+
+#if HAVE_SSSE3 && CONFIG_EXT_INTERP
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, VP10ConvolveOptimzTest,
+    ::testing::Combine(
+         ::testing::Values(vp10_convolve_horiz_ssse3),
+         ::testing::ValuesIn(kBlockDim),
+         ::testing::ValuesIn(kFilter),
+         ::testing::ValuesIn(kSubpelXQ4),
+         ::testing::ValuesIn(kAvg)));
+#endif  // HAVE_SSSE3 && CONFIG_EXT_INTERP
+}  // namespace

diff --git a/test/vp10_convolve_test.cc b/test/vp10_convolve_test.cc
index 42710a574fa25e87ad09868391294deb937f4f6f..901f578f6799c858cf62a7ef9610b544d5f1bb64 100644 (file)
--- a/test/vp10_convolve_test.cc
+++ b/test/vp10_convolve_test.cc
@@ -1,5 +1,6 @@
 #include "third_party/googletest/src/include/gtest/gtest.h"
 
+#include "./vp10_rtcd.h"
 #include "./vpx_dsp_rtcd.h"
 #include "test/acm_random.h"
 #include "vp10/common/filter.h"
@@ -40,6 +41,8 @@ TEST(VP10ConvolveTest, vp10_convolve8) {
   int w = 1;
   int h = 1;
 
+  vp10_rtcd();
+
   for (int i = 0; i < filter_size * filter_size; i++) {
     src[i] = rnd.Rand16() % (1 << 8);
   }
@@ -86,6 +89,8 @@ TEST(VP10ConvolveTest, vp10_convolve) {
   int subpel_x_q4;
   int subpel_y_q4;
 
+  vp10_rtcd();
+
   for (int i = 0; i < filter_size * filter_size; i++) {
     src[i] = rnd.Rand16() % (1 << 8);
   }
@@ -150,6 +155,8 @@ TEST(VP10ConvolveTest, vp10_convolve_avg) {
   int subpel_x_q4;
   int subpel_y_q4;
 
+  vp10_rtcd();
+
   for (int i = 0; i < filter_size * filter_size; i++) {
     src0[i] = rnd.Rand16() % (1 << 8);
     src1[i] = rnd.Rand16() % (1 << 8);

diff --git a/vp10/common/filter.c b/vp10/common/filter.c
index 154dade83ec2d71462902c978c706a8135b53cee..5dde3abd59d9542257be3b7c93488e49925a454b 100644 (file)
--- a/vp10/common/filter.c
+++ b/vp10/common/filter.c
@@ -302,3 +302,43 @@ const int16_t *vp10_get_interp_filter_kernel(
   return (const int16_t*)
       vp10_interp_filter_params_list[interp_filter].filter_ptr;
 }
+
+SubpelFilterCoeffs vp10_get_subpel_filter_signal_dir(
+    const InterpFilterParams p, int index) {
+#if CONFIG_EXT_INTERP && HAVE_SSSE3
+  if (p.filter_ptr == (const int16_t *)sub_pel_filters_12sharp) {
+    return &sub_pel_filters_12sharp_signal_dir[index][0];
+  }
+  if (p.filter_ptr == (const int16_t *)sub_pel_filters_10sharp) {
+    return &sub_pel_filters_10sharp_signal_dir[index][0];
+  }
+#endif
+#if USE_TEMPORALFILTER_12TAP && HAVE_SSSE3
+  if (p.filter_ptr == (const int16_t *)sub_pel_filters_temporalfilter_12) {
+    return &sub_pel_filters_temporalfilter_12_signal_dir[index][0];
+  }
+#endif
+  (void)p;
+  (void)index;
+  return NULL;
+}
+
+SubpelFilterCoeffs vp10_get_subpel_filter_ver_signal_dir(
+    const InterpFilterParams p, int index) {
+#if CONFIG_EXT_INTERP && HAVE_SSSE3
+  if (p.filter_ptr == (const int16_t *)sub_pel_filters_12sharp) {
+    return &sub_pel_filters_12sharp_ver_signal_dir[index][0];
+  }
+  if (p.filter_ptr == (const int16_t *)sub_pel_filters_10sharp) {
+    return &sub_pel_filters_10sharp_ver_signal_dir[index][0];
+  }
+#endif
+#if USE_TEMPORALFILTER_12TAP && HAVE_SSSE3
+  if (p.filter_ptr == (const int16_t *)sub_pel_filters_temporalfilter_12) {
+    return &sub_pel_filters_temporalfilter_12_ver_signal_dir[index][0];
+  }
+#endif
+  (void)p;
+  (void)index;
+  return NULL;
+}

diff --git a/vp10/common/filter.h b/vp10/common/filter.h
index a51e2d09d5e27e564e9e851a9ffd5cc843c1d721..591ac4d498651d9c14edbad0faf20256672dbb8a 100644 (file)
--- a/vp10/common/filter.h
+++ b/vp10/common/filter.h
@@ -91,6 +91,27 @@ static INLINE int vp10_is_interpolating_filter(
   const InterpFilterParams ip = vp10_get_interp_filter_params(interp_filter);
   return (ip.filter_ptr[ip.taps / 2 - 1] == 128);
 }
+
+#if USE_TEMPORALFILTER_12TAP
+extern const int8_t sub_pel_filters_temporalfilter_12_signal_dir[15][2][16];
+extern const int8_t sub_pel_filters_temporalfilter_12_ver_signal_dir[15][6][16];
+#endif
+
+#if CONFIG_EXT_INTERP
+extern const int8_t sub_pel_filters_12sharp_signal_dir[15][2][16];
+extern const int8_t sub_pel_filters_10sharp_signal_dir[15][2][16];
+extern const int8_t sub_pel_filters_12sharp_ver_signal_dir[15][6][16];
+extern const int8_t sub_pel_filters_10sharp_ver_signal_dir[15][6][16];
+#endif
+
+typedef const int8_t (*SubpelFilterCoeffs)[16];
+
+SubpelFilterCoeffs vp10_get_subpel_filter_signal_dir(
+    const InterpFilterParams p, int index);
+
+SubpelFilterCoeffs vp10_get_subpel_filter_ver_signal_dir(
+    const InterpFilterParams p, int index);
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif

diff --git a/vp10/common/vp10_convolve.c b/vp10/common/vp10_convolve.c
index d7e2eaf35ebd29e27c5bccf68282b840b3429c4d..b01653142de8a644e9538151afce83d048ae961f 100644 (file)
--- a/vp10/common/vp10_convolve.c
+++ b/vp10/common/vp10_convolve.c
@@ -1,6 +1,7 @@
 #include <assert.h>
 #include <string.h>
 
+#include "./vp10_rtcd.h"
 #include "vp10/common/filter.h"
 #include "vpx_dsp/vpx_dsp_common.h"
 #include "vpx_ports/mem.h"
@@ -10,7 +11,7 @@
 #define MAX_STEP (32)
 #define MAX_FILTER_TAP (12)
 
-static void convolve_horiz(const uint8_t *src, int src_stride, uint8_t *dst,
+void vp10_convolve_horiz_c(const uint8_t *src, int src_stride, uint8_t *dst,
                            int dst_stride, int w, int h,
                            const InterpFilterParams filter_params,
                            const int subpel_x_q4, int x_step_q4, int avg) {
@@ -121,8 +122,8 @@ void vp10_convolve(const uint8_t *src, int src_stride, uint8_t *dst,
         vp10_get_interp_filter_params(interp_filter);
 #endif
     assert(filter_params.taps <= MAX_FILTER_TAP);
-    convolve_horiz(src, src_stride, dst, dst_stride, w, h, filter_params,
-                   subpel_x_q4, x_step_q4, ref_idx);
+    vp10_convolve_horiz(src, src_stride, dst, dst_stride, w, h, filter_params,
+                        subpel_x_q4, x_step_q4, ref_idx);
   } else if (ignore_horiz) {
 #if CONFIG_DUAL_FILTER
     InterpFilterParams filter_params =
@@ -162,9 +163,9 @@ void vp10_convolve(const uint8_t *src, int src_stride, uint8_t *dst,
 
     assert(filter_params.taps <= MAX_FILTER_TAP);
 
-    convolve_horiz(src - src_stride * (filter_size / 2 - 1), src_stride, temp,
-                   temp_stride, w, intermediate_height, filter_params,
-                   subpel_x_q4, x_step_q4, 0);
+    vp10_convolve_horiz(src - src_stride * (filter_size / 2 - 1), src_stride,
+                        temp, temp_stride, w, intermediate_height,
+                        filter_params, subpel_x_q4, x_step_q4, 0);
 
 #if CONFIG_DUAL_FILTER
     filter_params = filter_params_y;

diff --git a/vp10/common/vp10_rtcd_defs.pl b/vp10/common/vp10_rtcd_defs.pl
index b30953d4612fdf0cf8f441043bb1eeefa198eea4..2942c5468cbe78bb9b90288b70b820940c04984d 100644 (file)
--- a/vp10/common/vp10_rtcd_defs.pl
+++ b/vp10/common/vp10_rtcd_defs.pl
@@ -8,6 +8,7 @@ print <<EOF
 #include "vp10/common/common.h"
 #include "vp10/common/enums.h"
 #include "vp10/common/quant_common.h"
+#include "vp10/common/filter.h"
 #include "vp10/common/vp10_txfm.h"
 
 struct macroblockd;
@@ -83,6 +84,12 @@ add_proto qw/void vp10_filter_by_weight8x8/, "const uint8_t *src, int src_stride
 specialize qw/vp10_filter_by_weight8x8 sse2 msa/;
 }
 
+#
+# 10/12-tap convolution filters
+#
+add_proto qw/void vp10_convolve_horiz/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, int avg";
+specialize qw/vp10_convolve_horiz ssse3/;
+
 #
 # dct
 #

diff --git a/vp10/common/x86/vp10_convolve_filters_ssse3.c b/vp10/common/x86/vp10_convolve_filters_ssse3.c
new file mode 100644 (file)
index 0000000..3be79ff
--- /dev/null
+++ b/vp10/common/x86/vp10_convolve_filters_ssse3.c
@@ -0,0 +1,632 @@
+/*
+ *  Copyright (c) 2016 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_config.h"
+#include "vp10/common/filter.h"
+
+// Note:
+//  Filter coefficients are from "filter.c".  We use,
+//  sub_pel_filters_temporalfilter_12[],
+//  sub_pel_filters_12sharp[],
+//  sub_pel_filters_10sharp[].
+
+// (2-1) Parallel filtering along the intended signal direction
+
+// 12-tap filter padding:
+// {filter_coefficients, 0, 0, 0, 0},
+// {0, 0, filter_coefficients, 0, 0),
+#if USE_TEMPORALFILTER_12TAP
+DECLARE_ALIGNED(16, const int8_t,
+                sub_pel_filters_temporalfilter_12_signal_dir[15][2][16]) = {
+  {
+    {0,   1,  -1,   3,  -7, 127,   8,  -4,   2,  -1,   0, 0, 0, 0, 0, 0},
+    {0, 0, 0,   1,  -1,   3,  -7, 127,   8,  -4,   2,  -1,   0, 0, 0, 0},
+  },
+  {
+    {0,   1,  -3,   5, -12, 124,  18,  -8,   4,  -2,   1, 0, 0, 0, 0, 0},
+    {0, 0, 0,   1,  -3,   5, -12, 124,  18,  -8,   4,  -2,   1, 0, 0, 0},
+  },
+  {
+    {-1,   2,  -4,   8, -17, 120,  28, -11,   6,  -3,   1, -1, 0, 0, 0, 0},
+    {0, 0, -1,   2,  -4,   8, -17, 120,  28, -11,   6,  -3,   1, -1, 0, 0},
+  },
+  {
+    {-1,   2,  -4,  10, -21, 114,  38, -15,   8,  -4,   2, -1, 0, 0, 0, 0},
+    {0, 0, -1,   2,  -4,  10, -21, 114,  38, -15,   8,  -4,   2, -1, 0, 0},
+  },
+  {
+    {-1,   3,  -5,  11, -23, 107,  49, -18,   9,  -5,   2, -1, 0, 0, 0, 0},
+    {0, 0, -1,   3,  -5,  11, -23, 107,  49, -18,   9,  -5,   2, -1, 0, 0},
+  },
+  {
+    {-1,   3,  -6,  12, -25,  99,  60, -21,  11,  -6,   3, -1, 0, 0, 0, 0},
+    {0, 0, -1,   3,  -6,  12, -25,  99,  60, -21,  11,  -6,   3, -1, 0, 0},
+  },
+  {
+    {-1,   3,  -6,  12, -25,  90,  70, -23,  12,  -6,   3, -1, 0, 0, 0, 0},
+    {0, 0, -1,   3,  -6,  12, -25,  90,  70, -23,  12,  -6,   3, -1, 0, 0},
+  },
+  {
+    {-1,   3,  -6,  12, -24,  80,  80, -24,  12,  -6,   3, -1, 0, 0, 0, 0},
+    {0, 0, -1,   3,  -6,  12, -24,  80,  80, -24,  12,  -6,   3, -1, 0, 0},
+  },
+  {
+    {-1,   3,  -6,  12, -23,  70,  90, -25,  12,  -6,   3, -1, 0, 0, 0, 0},
+    {0, 0, -1,   3,  -6,  12, -23,  70,  90, -25,  12,  -6,   3, -1, 0, 0},
+  },
+  {
+    {-1,   3,  -6,  11, -21,  60,  99, -25,  12,  -6,   3, -1, 0, 0, 0, 0},
+    {0, 0, -1,   3,  -6,  11, -21,  60,  99, -25,  12,  -6,   3, -1, 0, 0},
+  },
+  {
+    {-1,   2,  -5,   9, -18,  49, 107, -23,  11,  -5,   3, -1, 0, 0, 0, 0},
+    {0, 0, -1,   2,  -5,   9, -18,  49, 107, -23,  11,  -5,   3, -1, 0, 0},
+  },
+  {
+    {-1,   2,  -4,   8, -15,  38, 114, -21,  10,  -4,   2, -1, 0, 0, 0, 0},
+    {0, 0, -1,   2,  -4,   8, -15,  38, 114, -21,  10,  -4,   2, -1, 0, 0},
+  },
+  {
+    {-1,   1,  -3,   6, -11,  28, 120, -17,   8,  -4,   2, -1, 0, 0, 0, 0},
+    {0, 0, -1,   1,  -3,   6, -11,  28, 120, -17,   8,  -4,   2, -1, 0, 0},
+  },
+  {
+    {0,   1,  -2,   4,  -8,  18, 124, -12,   5,  -3,   1, 0, 0, 0, 0, 0},
+    {0, 0, 0,   1,  -2,   4,  -8,  18, 124, -12,   5,  -3,   1, 0, 0, 0},
+  },
+  {
+    {0,   0,  -1,   2,  -4,   8, 127,  -7,   3,  -1,   1, 0, 0, 0, 0, 0},
+    {0, 0, 0,   0,  -1,   2,  -4,   8, 127,  -7,   3,  -1,   1, 0, 0, 0},
+  },
+};
+#endif  // USE_TEMPORALFILTER_12TAP
+
+#if CONFIG_EXT_INTERP
+DECLARE_ALIGNED(16, const int8_t,
+                sub_pel_filters_12sharp_signal_dir[15][2][16]) = {
+  {
+    {0,   1,  -2,   3,  -7, 127,   8,  -4,   2,  -1,   1, 0, 0, 0, 0, 0},
+    {0, 0, 0,   1,  -2,   3,  -7, 127,   8,  -4,   2,  -1,   1, 0, 0, 0},
+  },
+  {
+    {-1,   2,  -3,   6, -13, 124,  18,  -8,   4,  -2,   2, -1, 0, 0, 0, 0},
+    {0, 0, -1,   2,  -3,   6, -13, 124,  18,  -8,   4,  -2,   2, -1, 0, 0},
+  },
+  {
+    {-1,   3,  -4,   8, -18, 120,  28, -12,   7,  -4,   2, -1, 0, 0, 0, 0},
+    {0, 0, -1,   3,  -4,   8, -18, 120,  28, -12,   7,  -4,   2, -1, 0, 0},
+  },
+  {
+    {-1,   3,  -6,  10, -21, 115,  38, -15,   8,  -5,   3, -1, 0, 0, 0, 0},
+    {0, 0, -1,   3,  -6,  10, -21, 115,  38, -15,   8,  -5,   3, -1, 0, 0},
+  },
+  {
+    {-2,   4,  -6,  12, -24, 108,  49, -18,  10,  -6,   3, -2, 0, 0, 0, 0},
+    {0, 0, -2,   4,  -6,  12, -24, 108,  49, -18,  10,  -6,   3, -2, 0, 0},
+  },
+  {
+    {-2,   4,  -7,  13, -25, 100,  60, -21,  11,  -7,   4, -2, 0, 0, 0, 0},
+    {0, 0, -2,   4,  -7,  13, -25, 100,  60, -21,  11,  -7,   4, -2, 0, 0},
+  },
+  {
+    {-2,   4,  -7,  13, -26,  91,  71, -24,  13,  -7,   4, -2, 0, 0, 0, 0},
+    {0, 0, -2,   4,  -7,  13, -26,  91,  71, -24,  13,  -7,   4, -2, 0, 0},
+  },
+  {
+    {-2,   4,  -7,  13, -25,  81,  81, -25,  13,  -7,   4, -2, 0, 0, 0, 0},
+    {0, 0, -2,   4,  -7,  13, -25,  81,  81, -25,  13,  -7,   4, -2, 0, 0},
+  },
+  {
+    {-2,   4,  -7,  13, -24,  71,  91, -26,  13,  -7,   4, -2, 0, 0, 0, 0},
+    {0, 0, -2,   4,  -7,  13, -24,  71,  91, -26,  13,  -7,   4, -2, 0, 0},
+  },
+  {
+    {-2,   4,  -7,  11, -21,  60, 100, -25,  13,  -7,   4, -2, 0, 0, 0, 0},
+    {0, 0, -2,   4,  -7,  11, -21,  60, 100, -25,  13,  -7,   4, -2, 0, 0},
+  },
+  {
+    {-2,   3,  -6,  10, -18,  49, 108, -24,  12,  -6,   4, -2, 0, 0, 0, 0},
+    {0, 0, -2,   3,  -6,  10, -18,  49, 108, -24,  12,  -6,   4, -2, 0, 0},
+  },
+  {
+    {-1,   3,  -5,   8, -15,  38, 115, -21,  10,  -6,   3, -1, 0, 0, 0, 0},
+    {0, 0, -1,   3,  -5,   8, -15,  38, 115, -21,  10,  -6,   3, -1, 0, 0},
+  },
+  {
+    {-1,   2,  -4,   7, -12,  28, 120, -18,   8,  -4,   3, -1, 0, 0, 0, 0},
+    {0, 0, -1,   2,  -4,   7, -12,  28, 120, -18,   8,  -4,   3, -1, 0, 0},
+  },
+  {
+    {-1,   2,  -2,   4,  -8,  18, 124, -13,   6,  -3,   2, -1, 0, 0, 0, 0},
+    {0, 0, -1,   2,  -2,   4,  -8,  18, 124, -13,   6,  -3,   2, -1, 0, 0},
+  },
+  {
+    {0,   1,  -1,   2,  -4,   8, 127,  -7,   3,  -2,   1, 0, 0, 0, 0, 0},
+    {0, 0, 0,   1,  -1,   2,  -4,   8, 127,  -7,   3,  -2,   1, 0, 0, 0},
+  },
+};
+
+// 10-tap filter padding:
+// {0, filter_coefficients, 0, 0, 0, 0, 0},
+// {0, 0, 0, filter_coefficients, 0, 0, 0),
+DECLARE_ALIGNED(16, const int8_t,
+                sub_pel_filters_10sharp_signal_dir[15][2][16]) = {
+  {
+    {0, 0,  -1,   3,  -6, 127,   8,  -4,   2,  -1, 0, 0, 0, 0, 0, 0},
+    {0, 0, 0,  -1,   3,  -6, 127,   8,  -4,   2,  -1, 0, 0, 0, 0},
+  },
+  {
+    {0, 1,  -2,   5, -12, 124,  18,  -7,   3,  -2, 0, 0, 0, 0, 0, 0},
+    {0, 0, 1,  -2,   5, -12, 124,  18,  -7,   3,  -2, 0, 0, 0, 0},
+  },
+  {
+    {0, 1,  -3,   7, -17, 119,  28, -11,   5,  -2, 1, 0, 0, 0, 0, 0},
+    {0, 0, 1,  -3,   7, -17, 119,  28, -11,   5,  -2, 1, 0, 0, 0},
+  },
+  {
+    {0, 1,  -4,   8, -20, 114,  38, -14,   7,  -3, 1, 0, 0, 0, 0, 0},
+    {0, 0, 1,  -4,   8, -20, 114,  38, -14,   7,  -3, 1, 0, 0, 0},
+  },
+  {
+    {0, 1,  -4,   9, -22, 107,  49, -17,   8,  -4, 1, 0, 0, 0, 0, 0},
+    {0, 0, 1,  -4,   9, -22, 107,  49, -17,   8,  -4, 1, 0, 0, 0},
+  },
+  {
+    {0, 2,  -5,  10, -24,  99,  59, -20,   9,  -4, 2, 0, 0, 0, 0, 0},
+    {0, 0, 2,  -5,  10, -24,  99,  59, -20,   9,  -4, 2, 0, 0, 0},
+  },
+  {
+    {0, 2,  -5,  10, -24,  90,  70, -22,  10,  -5, 2, 0, 0, 0, 0, 0},
+    {0, 0, 2,  -5,  10, -24,  90,  70, -22,  10,  -5, 2, 0, 0, 0},
+  },
+  {
+    {0, 2,  -5,  10, -23,  80,  80, -23,  10,  -5, 2, 0, 0, 0, 0, 0},
+    {0, 0, 2,  -5,  10, -23,  80,  80, -23,  10,  -5, 2, 0, 0, 0},
+  },
+  {
+    {0, 2,  -5,  10, -22,  70,  90, -24,  10,  -5, 2, 0, 0, 0, 0, 0},
+    {0, 0, 2,  -5,  10, -22,  70,  90, -24,  10,  -5, 2, 0, 0, 0},
+  },
+  {
+    {0, 2,  -4,   9, -20,  59,  99, -24,  10,  -5, 2, 0, 0, 0, 0, 0},
+    {0, 0, 2,  -4,   9, -20,  59,  99, -24,  10,  -5, 2, 0, 0, 0},
+  },
+  {
+    {0, 1,  -4,   8, -17,  49, 107, -22,   9,  -4, 1, 0, 0, 0, 0, 0},
+    {0, 0, 1,  -4,   8, -17,  49, 107, -22,   9,  -4, 1, 0, 0, 0},
+  },
+  {
+    {0, 1,  -3,   7, -14,  38, 114, -20,   8,  -4, 1, 0, 0, 0, 0, 0},
+    {0, 0, 1,  -3,   7, -14,  38, 114, -20,   8,  -4, 1, 0, 0, 0},
+  },
+  {
+    {0, 1,  -2,   5, -11,  28, 119, -17,   7,  -3, 1, 0, 0, 0, 0, 0},
+    {0, 0, 1,  -2,   5, -11,  28, 119, -17,   7,  -3, 1, 0, 0, 0},
+  },
+  {
+    {0, 0,  -2,   3,  -7,  18, 124, -12,   5,  -2, 1, 0, 0, 0, 0, 0},
+    {0, 0, 0,  -2,   3,  -7,  18, 124, -12,   5,  -2, 1, 0, 0, 0},
+  },
+  {
+    {0, 0,  -1,   2,  -4,   8, 127,  -6,   3,  -1, 0, 0, 0, 0, 0, 0},
+    {0, 0, 0,  -1,   2,  -4,   8, 127,  -6,   3,  -1, 0, 0, 0, 0},
+  },
+};
+#endif  // CONFIG_EXT_INTERP
+
+// (2-2) Parallel filtering vertically to signal direction
+#if USE_TEMPORALFILTER_12TAP
+DECLARE_ALIGNED(16, const int8_t,
+                sub_pel_filters_temporalfilter_12_ver_signal_dir[15][6][16]) = {
+  {
+    {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1},
+    {-1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3},
+    {-7, 127, -7, 127, -7, 127, -7, 127, -7, 127, -7, 127,
+     -7, 127, -7, 127},
+    {8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4},
+    {2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1},
+    {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
+  },
+  {
+    {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1},
+    {-3, 5, -3, 5, -3, 5, -3, 5, -3, 5, -3, 5, -3, 5, -3, 5},
+    {-12, 124, -12, 124, -12, 124, -12, 124, -12, 124, -12, 124,
+     -12, 124, -12, 124},
+    {18, -8, 18, -8, 18, -8, 18, -8, 18, -8, 18, -8, 18, -8, 18, -8},
+    {4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2},
+    {1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0},
+  },
+  {
+    {-1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2},
+    {-4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8},
+    {-17, 120, -17, 120, -17, 120, -17, 120, -17, 120, -17, 120,
+     -17, 120, -17, 120},
+    {28, -11, 28, -11, 28, -11, 28, -11, 28, -11, 28, -11, 28, -11, 28, -11},
+    {6, -3, 6, -3, 6, -3, 6, -3, 6, -3, 6, -3, 6, -3, 6, -3},
+    {1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1},
+  },
+  {
+    {-1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2},
+    {-4, 10, -4, 10, -4, 10, -4, 10, -4, 10, -4, 10, -4, 10, -4, 10},
+    {-21, 114, -21, 114, -21, 114, -21, 114, -21, 114, -21, 114,
+     -21, 114, -21, 114},
+    {38, -15, 38, -15, 38, -15, 38, -15, 38, -15, 38, -15,
+     38, -15, 38, -15},
+    {8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4},
+    {2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1},
+  },
+  {
+    {-1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3},
+    {-5, 11, -5, 11, -5, 11, -5, 11, -5, 11, -5, 11, -5, 11, -5, 11},
+    {-23, 107, -23, 107, -23, 107, -23, 107, -23, 107, -23, 107,
+     -23, 107, -23, 107},
+    {49, -18, 49, -18, 49, -18, 49, -18, 49, -18, 49, -18, 49, -18, 49, -18},
+    {9, -5, 9, -5, 9, -5, 9, -5, 9, -5, 9, -5, 9, -5, 9, -5},
+    {2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1},
+  },
+  {
+    {-1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3},
+    {-6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12},
+    {-25, 99, -25, 99, -25, 99, -25, 99, -25, 99, -25, 99, -25, 99, -25, 99},
+    {60, -21, 60, -21, 60, -21, 60, -21, 60, -21, 60, -21, 60, -21, 60, -21},
+    {11, -6, 11, -6, 11, -6, 11, -6, 11, -6, 11, -6, 11, -6, 11, -6},
+    {3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1},
+  },
+  {
+    {-1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3},
+    {-6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12},
+    {-25, 90, -25, 90, -25, 90, -25, 90, -25, 90, -25, 90, -25, 90, -25, 90},
+    {70, -23, 70, -23, 70, -23, 70, -23, 70, -23, 70, -23, 70, -23, 70, -23},
+    {12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6},
+    {3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1},
+  },
+  {
+    {-1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3},
+    {-6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12},
+    {-24, 80, -24, 80, -24, 80, -24, 80, -24, 80, -24, 80, -24, 80, -24, 80},
+    {80, -24, 80, -24, 80, -24, 80, -24, 80, -24, 80, -24, 80, -24, 80, -24},
+    {12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6},
+    {3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1},
+  },
+  {
+    {-1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3},
+    {-6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12},
+    {-23, 70, -23, 70, -23, 70, -23, 70, -23, 70, -23, 70, -23, 70, -23, 70},
+    {90, -25, 90, -25, 90, -25, 90, -25, 90, -25, 90, -25, 90, -25, 90, -25},
+    {12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6},
+    {3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1},
+  },
+  {
+    {-1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3},
+    {-6, 11, -6, 11, -6, 11, -6, 11, -6, 11, -6, 11, -6, 11, -6, 11},
+    {-21, 60, -21, 60, -21, 60, -21, 60, -21, 60, -21, 60, -21, 60, -21, 60},
+    {99, -25, 99, -25, 99, -25, 99, -25, 99, -25, 99, -25, 99, -25, 99, -25},
+    {12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6},
+    {3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1},
+  },
+  {
+    {-1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2},
+    {-5, 9, -5, 9, -5, 9, -5, 9, -5, 9, -5, 9, -5, 9, -5, 9},
+    {-18, 49, -18, 49, -18, 49, -18, 49, -18, 49, -18, 49, -18, 49, -18, 49},
+    {107, -23, 107, -23, 107, -23, 107, -23, 107, -23, 107, -23, 107,
+     -23, 107, -23},
+    {11, -5, 11, -5, 11, -5, 11, -5, 11, -5, 11, -5, 11, -5, 11, -5},
+    {3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1},
+  },
+  {
+    {-1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2},
+    {-4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8},
+    {-15, 38, -15, 38, -15, 38, -15, 38, -15, 38, -15, 38, -15, 38, -15, 38},
+    {114, -21, 114, -21, 114, -21, 114, -21, 114, -21, 114, -21,
+     114, -21, 114, -21},
+    {10, -4, 10, -4, 10, -4, 10, -4, 10, -4, 10, -4, 10, -4, 10, -4},
+    {2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1},
+  },
+  {
+    {-1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1, -1, 1},
+    {-3, 6, -3, 6, -3, 6, -3, 6, -3, 6, -3, 6, -3, 6, -3, 6},
+    {-11, 28, -11, 28, -11, 28, -11, 28, -11, 28, -11, 28, -11, 28, -11, 28},
+    {120, -17, 120, -17, 120, -17, 120, -17, 120, -17, 120, -17,
+     120, -17, 120, -17},
+    {8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4},
+    {2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1},
+  },
+  {
+    {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1},
+    {-2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4},
+    {-8, 18, -8, 18, -8, 18, -8, 18, -8, 18, -8, 18, -8, 18, -8, 18},
+    {124, -12, 124, -12, 124, -12, 124, -12, 124, -12, 124, -12,
+     124, -12, 124, -12},
+    {5, -3, 5, -3, 5, -3, 5, -3, 5, -3, 5, -3, 5, -3, 5, -3},
+    {1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0},
+  },
+  {
+    {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
+    {-1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2},
+    {-4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8},
+    {127, -7, 127, -7, 127, -7, 127, -7, 127, -7, 127, -7,
+     127, -7, 127, -7},
+    {3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1},
+    {1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0},
+  },
+};
+#endif  // USE_TEMPORALFILTER_12TAP
+
+#if CONFIG_EXT_INTERP
+DECLARE_ALIGNED(16, const int8_t,
+                sub_pel_filters_12sharp_ver_signal_dir[15][6][16]) = {
+  {
+    {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1},
+    {-2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3},
+    {-7, 127, -7, 127, -7, 127, -7, 127, -7, 127, -7, 127,
+     -7, 127, -7, 127},
+    {8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4},
+    {2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1},
+    {1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0},
+  },
+  {
+    {-1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2},
+    {-3, 6, -3, 6, -3, 6, -3, 6, -3, 6, -3, 6, -3, 6, -3, 6},
+    {-13, 124, -13, 124, -13, 124, -13, 124, -13, 124, -13, 124,
+     -13, 124, -13, 124},
+    {18, -8, 18, -8, 18, -8, 18, -8, 18, -8, 18, -8, 18, -8, 18, -8},
+    {4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2},
+    {2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1},
+  },
+  {
+    {-1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3},
+    {-4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8},
+    {-18, 120, -18, 120, -18, 120, -18, 120, -18, 120, -18, 120,
+     -18, 120, -18, 120},
+    {28, -12, 28, -12, 28, -12, 28, -12, 28, -12, 28, -12, 28, -12, 28, -12},
+    {7, -4, 7, -4, 7, -4, 7, -4, 7, -4, 7, -4, 7, -4, 7, -4},
+    {2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1},
+  },
+  {
+    {-1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3},
+    {-6, 10, -6, 10, -6, 10, -6, 10, -6, 10, -6, 10, -6, 10, -6, 10},
+    {-21, 115, -21, 115, -21, 115, -21, 115, -21, 115, -21, 115,
+     -21, 115, -21, 115},
+    {38, -15, 38, -15, 38, -15, 38, -15, 38, -15, 38, -15,
+     38, -15, 38, -15},
+    {8, -5, 8, -5, 8, -5, 8, -5, 8, -5, 8, -5, 8, -5, 8, -5},
+    {3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1},
+  },
+  {
+    {-2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4},
+    {-6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12},
+    {-24, 108, -24, 108, -24, 108, -24, 108, -24, 108, -24, 108,
+     -24, 108, -24, 108},
+    {49, -18, 49, -18, 49, -18, 49, -18, 49, -18, 49, -18, 49, -18, 49, -18},
+    {10, -6, 10, -6, 10, -6, 10, -6, 10, -6, 10, -6, 10, -6, 10, -6},
+    {3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2},
+  },
+  {
+    {-2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4},
+    {-7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13},
+    {-25, 100, -25, 100, -25, 100, -25, 100, -25, 100, -25, 100,
+     -25, 100, -25, 100},
+    {60, -21, 60, -21, 60, -21, 60, -21, 60, -21, 60, -21, 60, -21, 60, -21},
+    {11, -7, 11, -7, 11, -7, 11, -7, 11, -7, 11, -7, 11, -7, 11, -7},
+    {4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2},
+  },
+  {
+    {-2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4},
+    {-7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13},
+    {-26, 91, -26, 91, -26, 91, -26, 91, -26, 91, -26, 91, -26, 91, -26, 91},
+    {71, -24, 71, -24, 71, -24, 71, -24, 71, -24, 71, -24, 71, -24, 71, -24},
+    {13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7},
+    {4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2},
+  },
+  {
+    {-2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4},
+    {-7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13},
+    {-25, 81, -25, 81, -25, 81, -25, 81, -25, 81, -25, 81, -25, 81, -25, 81},
+    {81, -25, 81, -25, 81, -25, 81, -25, 81, -25, 81, -25, 81, -25, 81, -25},
+    {13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7},
+    {4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2},
+  },
+  {
+    {-2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4},
+    {-7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13},
+    {-24, 71, -24, 71, -24, 71, -24, 71, -24, 71, -24, 71, -24, 71, -24, 71},
+    {91, -26, 91, -26, 91, -26, 91, -26, 91, -26, 91, -26, 91, -26, 91, -26},
+    {13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7},
+    {4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2},
+  },
+  {
+    {-2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4},
+    {-7, 11, -7, 11, -7, 11, -7, 11, -7, 11, -7, 11, -7, 11, -7, 11},
+    {-21, 60, -21, 60, -21, 60, -21, 60, -21, 60, -21, 60, -21, 60, -21, 60},
+    {100, -25, 100, -25, 100, -25, 100, -25, 100, -25, 100, -25,
+     100, -25, 100, -25},
+    {13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7, 13, -7},
+    {4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2},
+  },
+  {
+    {-2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3},
+    {-6, 10, -6, 10, -6, 10, -6, 10, -6, 10, -6, 10, -6, 10, -6, 10},
+    {-18, 49, -18, 49, -18, 49, -18, 49, -18, 49, -18, 49, -18, 49, -18, 49},
+    {108, -24, 108, -24, 108, -24, 108, -24, 108, -24, 108, -24,
+     108, -24, 108, -24},
+    {12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6, 12, -6},
+    {4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2},
+  },
+  {
+    {-1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3},
+    {-5, 8, -5, 8, -5, 8, -5, 8, -5, 8, -5, 8, -5, 8, -5, 8},
+    {-15, 38, -15, 38, -15, 38, -15, 38, -15, 38, -15, 38, -15, 38, -15, 38},
+    {115, -21, 115, -21, 115, -21, 115, -21, 115, -21, 115, -21,
+     115, -21, 115, -21},
+    {10, -6, 10, -6, 10, -6, 10, -6, 10, -6, 10, -6, 10, -6, 10, -6},
+    {3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1},
+  },
+  {
+    {-1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2},
+    {-4, 7, -4, 7, -4, 7, -4, 7, -4, 7, -4, 7, -4, 7, -4, 7},
+    {-12, 28, -12, 28, -12, 28, -12, 28, -12, 28, -12, 28, -12, 28, -12, 28},
+    {120, -18, 120, -18, 120, -18, 120, -18, 120, -18, 120, -18,
+     120, -18, 120, -18},
+    {8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4},
+    {3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1},
+  },
+  {
+    {-1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2},
+    {-2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4, -2, 4},
+    {-8, 18, -8, 18, -8, 18, -8, 18, -8, 18, -8, 18, -8, 18, -8, 18},
+    {124, -13, 124, -13, 124, -13, 124, -13, 124, -13, 124, -13,
+     124, -13, 124, -13},
+    {6, -3, 6, -3, 6, -3, 6, -3, 6, -3, 6, -3, 6, -3, 6, -3},
+    {2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1},
+  },
+  {
+    {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1},
+    {-1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2},
+    {-4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8},
+    {127, -7, 127, -7, 127, -7, 127, -7, 127, -7, 127, -7,
+     127, -7, 127, -7},
+    {3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2},
+    {1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0},
+  },
+};
+
+DECLARE_ALIGNED(16, const int8_t,
+                sub_pel_filters_10sharp_ver_signal_dir[15][6][16]) = {
+  {
+    {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
+    {-1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3},
+    {-6, 127, -6, 127, -6, 127, -6, 127, -6, 127, -6, 127, -6, 127, -6, 127},
+    {8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4},
+    {2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1},
+    {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
+  },
+  {
+    {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1},
+    {-2, 5, -2, 5, -2, 5, -2, 5, -2, 5, -2, 5, -2, 5, -2, 5},
+    {-12, 124, -12, 124, -12, 124, -12, 124, -12, 124, -12, 124,
+     -12, 124, -12, 124},
+    {18, -7, 18, -7, 18, -7, 18, -7, 18, -7, 18, -7, 18, -7, 18, -7},
+    {3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2},
+    {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
+  },
+  {
+    {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1},
+    {-3, 7, -3, 7, -3, 7, -3, 7, -3, 7, -3, 7, -3, 7, -3, 7},
+    {-17, 119, -17, 119, -17, 119, -17, 119, -17, 119, -17, 119,
+     -17, 119, -17, 119},
+    {28, -11, 28, -11, 28, -11, 28, -11, 28, -11, 28, -11, 28, -11, 28, -11},
+    {5, -2, 5, -2, 5, -2, 5, -2, 5, -2, 5, -2, 5, -2, 5, -2},
+    {1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0},
+  },
+  {
+    {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1},
+    {-4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8},
+    {-20, 114, -20, 114, -20, 114, -20, 114, -20, 114, -20, 114,
+     -20, 114, -20, 114},
+    {38, -14, 38, -14, 38, -14, 38, -14, 38, -14, 38, -14, 38, -14, 38, -14},
+    {7, -3, 7, -3, 7, -3, 7, -3, 7, -3, 7, -3, 7, -3, 7, -3},
+    {1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0},
+  },
+  {
+    {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1},
+    {-4, 9, -4, 9, -4, 9, -4, 9, -4, 9, -4, 9, -4, 9, -4, 9},
+    {-22, 107, -22, 107, -22, 107, -22, 107, -22, 107, -22, 107,
+     -22, 107, -22, 107},
+    {49, -17, 49, -17, 49, -17, 49, -17, 49, -17, 49, -17, 49, -17, 49, -17},
+    {8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4},
+    {1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0},
+  },
+  {
+    {0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2},
+    {-5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10},
+    {-24, 99, -24, 99, -24, 99, -24, 99, -24, 99, -24, 99, -24, 99, -24, 99},
+    {59, -20, 59, -20, 59, -20, 59, -20, 59, -20, 59, -20, 59, -20, 59, -20},
+    {9, -4, 9, -4, 9, -4, 9, -4, 9, -4, 9, -4, 9, -4, 9, -4},
+    {2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0},
+  },
+  {
+    {0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2},
+    {-5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10},
+    {-24, 90, -24, 90, -24, 90, -24, 90, -24, 90, -24, 90, -24, 90, -24, 90},
+    {70, -22, 70, -22, 70, -22, 70, -22, 70, -22, 70, -22, 70, -22, 70, -22},
+    {10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5},
+    {2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0},
+  },
+  {
+    {0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2},
+    {-5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10},
+    {-23, 80, -23, 80, -23, 80, -23, 80, -23, 80, -23, 80, -23, 80, -23, 80},
+    {80, -23, 80, -23, 80, -23, 80, -23, 80, -23, 80, -23, 80, -23, 80, -23},
+    {10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5},
+    {2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0},
+  },
+  {
+    {0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2},
+    {-5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10},
+    {-22, 70, -22, 70, -22, 70, -22, 70, -22, 70, -22, 70, -22, 70, -22, 70},
+    {90, -24, 90, -24, 90, -24, 90, -24, 90, -24, 90, -24, 90, -24, 90, -24},
+    {10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5},
+    {2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0},
+  },
+  {
+    {0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2},
+    {-4, 9, -4, 9, -4, 9, -4, 9, -4, 9, -4, 9, -4, 9, -4, 9},
+    {-20, 59, -20, 59, -20, 59, -20, 59, -20, 59, -20, 59, -20, 59, -20, 59},
+    {99, -24, 99, -24, 99, -24, 99, -24, 99, -24, 99, -24, 99, -24, 99, -24},
+    {10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5, 10, -5},
+    {2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0},
+  },
+  {
+    {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1},
+    {-4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8},
+    {-17, 49, -17, 49, -17, 49, -17, 49, -17, 49, -17, 49, -17, 49, -17, 49},
+    {107, -22, 107, -22, 107, -22, 107, -22, 107, -22, 107, -22,
+     107, -22, 107, -22},
+    {9, -4, 9, -4, 9, -4, 9, -4, 9, -4, 9, -4, 9, -4, 9, -4},
+    {1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0},
+  },
+  {
+    {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1},
+    {-3, 7, -3, 7, -3, 7, -3, 7, -3, 7, -3, 7, -3, 7, -3, 7},
+    {-14, 38, -14, 38, -14, 38, -14, 38, -14, 38, -14, 38, -14, 38, -14, 38},
+    {114, -20, 114, -20, 114, -20, 114, -20, 114, -20, 114, -20,
+     114, -20, 114, -20},
+    {8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4},
+    {1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0},
+  },
+  {
+    {0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1},
+    {-2, 5, -2, 5, -2, 5, -2, 5, -2, 5, -2, 5, -2, 5, -2, 5},
+    {-11, 28, -11, 28, -11, 28, -11, 28, -11, 28, -11, 28, -11, 28, -11, 28},
+    {119, -17, 119, -17, 119, -17, 119, -17, 119, -17, 119, -17,
+     119, -17, 119, -17},
+    {7, -3, 7, -3, 7, -3, 7, -3, 7, -3, 7, -3, 7, -3, 7, -3},
+    {1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0},
+  },
+  {
+    {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
+    {-2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3, -2, 3},
+    {-7, 18, -7, 18, -7, 18, -7, 18, -7, 18, -7, 18, -7, 18, -7, 18},
+    {124, -12, 124, -12, 124, -12, 124, -12, 124, -12, 124, -12,
+     124, -12, 124, -12},
+    {5, -2, 5, -2, 5, -2, 5, -2, 5, -2, 5, -2, 5, -2, 5, -2},
+    {1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0},
+  },
+  {
+    {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
+    {-1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2, -1, 2},
+    {-4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8, -4, 8},
+    {127, -6, 127, -6, 127, -6, 127, -6, 127, -6, 127, -6,
+     127, -6, 127, -6},
+    {3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1, 3, -1},
+    {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
+  },
+};
+#endif  // CONFIG_EXT_INTERP

diff --git a/vp10/common/x86/vp10_convolve_ssse3.c b/vp10/common/x86/vp10_convolve_ssse3.c
new file mode 100644 (file)
index 0000000..91bd2d4
--- /dev/null
+++ b/vp10/common/x86/vp10_convolve_ssse3.c
@@ -0,0 +1,716 @@
+/*
+ *  Copyright (c) 2016 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <tmmintrin.h>
+
+#include "./vp10_rtcd.h"
+#include "vp10/common/filter.h"
+
+#define WIDTH_BOUND  (16)
+#define HEIGHT_BOUND (16)
+
+static INLINE void transpose_4x8(const __m128i *in, __m128i *out) {
+  __m128i t0, t1;
+
+  t0 = _mm_unpacklo_epi16(in[0], in[1]);
+  t1 = _mm_unpacklo_epi16(in[2], in[3]);
+
+  out[0] = _mm_unpacklo_epi32(t0, t1);
+  out[1] = _mm_srli_si128(out[0], 8);
+  out[2] = _mm_unpackhi_epi32(t0, t1);
+  out[3] = _mm_srli_si128(out[2], 8);
+
+  t0 = _mm_unpackhi_epi16(in[0], in[1]);
+  t1 = _mm_unpackhi_epi16(in[2], in[3]);
+
+  out[4] = _mm_unpacklo_epi32(t0, t1);
+  out[5] = _mm_srli_si128(out[4], 8);
+  // Note: We ignore out[6] and out[7] because
+  // they're zero vectors.
+}
+
+typedef void (*store_pixel_t)(__m128i x, uint8_t *src, uint8_t *dst);
+
+static INLINE void store_4_pixel_only(__m128i x, uint8_t *src, uint8_t *dst) {
+  (void)src;
+  x = _mm_packus_epi16(x, x);
+  *(int *)dst = _mm_cvtsi128_si32(x);
+}
+
+static INLINE __m128i accumulate_store(__m128i x, uint8_t *src) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i one = _mm_set1_epi16(1);
+  __m128i y = _mm_loadl_epi64((__m128i const *)src);
+  y = _mm_unpacklo_epi8(y, zero);
+  y = _mm_add_epi16(x, y);
+  y = _mm_add_epi16(y, one);
+  y = _mm_srai_epi16(y, 1);
+  y = _mm_packus_epi16(y, y);
+  return y;
+}
+
+static INLINE void accumulate_store_4_pixel(__m128i x, uint8_t *src,
+                                            uint8_t *dst) {
+  __m128i y = accumulate_store(x, src);
+  *(int *)dst = _mm_cvtsi128_si32(y);
+}
+
+static store_pixel_t store4pixelTab[2] = {
+  store_4_pixel_only, accumulate_store_4_pixel};
+
+void horiz_w4_ssse3(const uint8_t *src, const __m128i *f,
+                    int tapsNum, store_pixel_t store_func, uint8_t *dst,
+                    uint8_t *buf) {
+  __m128i sumPairRow[4];
+  __m128i sumPairCol[8];
+  __m128i pixel;
+  const __m128i k_256 = _mm_set1_epi16(1 << 8);
+
+  if (10 == tapsNum) {
+    src -= 1;
+  }
+
+  pixel = _mm_loadu_si128((__m128i const *)src);
+  sumPairRow[0] = _mm_maddubs_epi16(pixel, f[0]);
+  sumPairRow[2] = _mm_maddubs_epi16(pixel, f[1]);
+  sumPairRow[2] = _mm_srli_si128(sumPairRow[2], 2);
+
+  pixel = _mm_loadu_si128((__m128i const *)(src + 1));
+  sumPairRow[1] = _mm_maddubs_epi16(pixel, f[0]);
+  sumPairRow[3] = _mm_maddubs_epi16(pixel, f[1]);
+  sumPairRow[3] = _mm_srli_si128(sumPairRow[3], 2);
+
+  transpose_4x8(sumPairRow, sumPairCol);
+
+  sumPairRow[0] = _mm_adds_epi16(sumPairCol[0], sumPairCol[1]);
+  sumPairRow[1] = _mm_adds_epi16(sumPairCol[4], sumPairCol[5]);
+
+  sumPairRow[2] = _mm_min_epi16(sumPairCol[2], sumPairCol[3]);
+  sumPairRow[3] = _mm_max_epi16(sumPairCol[2], sumPairCol[3]);
+
+  sumPairRow[0] = _mm_adds_epi16(sumPairRow[0], sumPairRow[1]);
+  sumPairRow[0] = _mm_adds_epi16(sumPairRow[0], sumPairRow[2]);
+  sumPairRow[0] = _mm_adds_epi16(sumPairRow[0], sumPairRow[3]);
+
+  sumPairRow[1] = _mm_mulhrs_epi16(sumPairRow[0], k_256);
+
+  store_func(sumPairRow[1], dst, buf);
+}
+
+void horiz_w8_ssse3(const uint8_t *src, const __m128i *f, int tapsNum,
+                    store_pixel_t store, uint8_t *dst, uint8_t *buf) {
+  horiz_w4_ssse3(src, f, tapsNum, store, dst, buf);
+  src += 4;
+  buf += 4;
+  horiz_w4_ssse3(src, f, tapsNum, store, dst, buf);
+}
+
+void horiz_w16_ssse3(const uint8_t *src, const __m128i *f, int tapsNum,
+                     store_pixel_t store, uint8_t *dst, uint8_t *buf) {
+  horiz_w8_ssse3(src, f, tapsNum, store, dst, buf);
+  src += 8;
+  buf += 8;
+  horiz_w8_ssse3(src, f, tapsNum, store, dst, buf);
+}
+
+void horiz_w32_ssse3(const uint8_t *src, const __m128i *f, int tapsNum,
+                     store_pixel_t store, uint8_t *dst, uint8_t *buf) {
+  horiz_w16_ssse3(src, f, tapsNum, store, dst, buf);
+  src += 16;
+  buf += 16;
+  horiz_w16_ssse3(src, f, tapsNum, store, dst, buf);
+}
+
+void horiz_w64_ssse3(const uint8_t *src, const __m128i *f, int tapsNum,
+                     store_pixel_t store, uint8_t *dst, uint8_t *buf) {
+  horiz_w32_ssse3(src, f, tapsNum, store, dst, buf);
+  src += 32;
+  buf += 32;
+  horiz_w32_ssse3(src, f, tapsNum, store, dst, buf);
+}
+
+void horiz_w128_ssse3(const uint8_t *src, const __m128i *f, int tapsNum,
+                      store_pixel_t store, uint8_t *dst, uint8_t *buf) {
+  horiz_w64_ssse3(src, f, tapsNum, store, dst, buf);
+  src += 64;
+  buf += 64;
+  horiz_w64_ssse3(src, f, tapsNum, store, dst, buf);
+}
+
+static void (*horizTab[6])(const uint8_t *, const __m128i *, int,
+                           store_pixel_t, uint8_t *, uint8_t *) = {
+  horiz_w4_ssse3,
+  horiz_w8_ssse3,
+  horiz_w16_ssse3,
+  horiz_w32_ssse3,
+  horiz_w64_ssse3,
+  horiz_w128_ssse3,
+};
+
+void filter_horiz_ssse3(const uint8_t *src, __m128i *f, int tapsNum, int width,
+                        store_pixel_t store, uint8_t *dst, uint8_t *buffer) {
+  switch (width) {
+    case 2:
+    case 4:
+      horizTab[0](src, f, tapsNum, store, dst, buffer);
+      break;
+    case 8:
+      horizTab[1](src, f, tapsNum, store, dst, buffer);
+      break;
+    case 16:
+      horizTab[2](src, f, tapsNum, store, dst, buffer);
+      break;
+    case 32:
+      horizTab[3](src, f, tapsNum, store, dst, buffer);
+      break;
+    case 64:
+      horizTab[4](src, f, tapsNum, store, dst, buffer);
+      break;
+    case 128:
+      horizTab[5](src, f, tapsNum, store, dst, buffer);
+      break;
+    default:
+      assert(0);
+  }
+}
+
+// Vertical 8-pixel parallel
+typedef void (*transpose_to_dst_t)(const uint16_t *src, int src_stride,
+                                   uint8_t *dst, int dst_stride);
+
+static INLINE void transpose8x8_direct_to_dst(const uint16_t *src,
+                                              int src_stride,
+                                              uint8_t *dst,
+                                              int dst_stride) {
+  const __m128i k_256 = _mm_set1_epi16(1 << 8);
+  __m128i v0, v1, v2, v3;
+
+  __m128i u0 = _mm_loadu_si128((__m128i const *)(src + 0 * src_stride));
+  __m128i u1 = _mm_loadu_si128((__m128i const *)(src + 1 * src_stride));
+  __m128i u2 = _mm_loadu_si128((__m128i const *)(src + 2 * src_stride));
+  __m128i u3 = _mm_loadu_si128((__m128i const *)(src + 3 * src_stride));
+  __m128i u4 = _mm_loadu_si128((__m128i const *)(src + 4 * src_stride));
+  __m128i u5 = _mm_loadu_si128((__m128i const *)(src + 5 * src_stride));
+  __m128i u6 = _mm_loadu_si128((__m128i const *)(src + 6 * src_stride));
+  __m128i u7 = _mm_loadu_si128((__m128i const *)(src + 7 * src_stride));
+
+  u0 = _mm_mulhrs_epi16(u0, k_256);
+  u1 = _mm_mulhrs_epi16(u1, k_256);
+  u2 = _mm_mulhrs_epi16(u2, k_256);
+  u3 = _mm_mulhrs_epi16(u3, k_256);
+  u4 = _mm_mulhrs_epi16(u4, k_256);
+  u5 = _mm_mulhrs_epi16(u5, k_256);
+  u6 = _mm_mulhrs_epi16(u6, k_256);
+  u7 = _mm_mulhrs_epi16(u7, k_256);
+
+  v0 = _mm_packus_epi16(u0, u1);
+  v1 = _mm_packus_epi16(u2, u3);
+  v2 = _mm_packus_epi16(u4, u5);
+  v3 = _mm_packus_epi16(u6, u7);
+
+  u0 = _mm_unpacklo_epi8(v0, v1);
+  u1 = _mm_unpackhi_epi8(v0, v1);
+  u2 = _mm_unpacklo_epi8(v2, v3);
+  u3 = _mm_unpackhi_epi8(v2, v3);
+
+  u4 = _mm_unpacklo_epi8(u0, u1);
+  u5 = _mm_unpacklo_epi8(u2, u3);
+  u6 = _mm_unpackhi_epi8(u0, u1);
+  u7 = _mm_unpackhi_epi8(u2, u3);
+
+  u0 = _mm_unpacklo_epi32(u4, u5);
+  u1 = _mm_unpackhi_epi32(u4, u5);
+  u2 = _mm_unpacklo_epi32(u6, u7);
+  u3 = _mm_unpackhi_epi32(u6, u7);
+
+  u4 = _mm_srli_si128(u0, 8);
+  u5 = _mm_srli_si128(u1, 8);
+  u6 = _mm_srli_si128(u2, 8);
+  u7 = _mm_srli_si128(u3, 8);
+
+  _mm_storel_epi64((__m128i*)dst, u0);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 1), u4);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 2), u1);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 3), u5);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 4), u2);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 5), u6);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 6), u3);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 7), u7);
+}
+
+static INLINE void transpose8x8_accumu_to_dst(const uint16_t *src,
+                                              int src_stride,
+                                              uint8_t *dst,
+                                              int dst_stride) {
+  const __m128i k_256 = _mm_set1_epi16(1 << 8);
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i one = _mm_set1_epi16(1);
+  __m128i v0, v1, v2, v3, v4, v5, v6, v7;
+
+  __m128i u0 = _mm_loadu_si128((__m128i const *)(src + 0 * src_stride));
+  __m128i u1 = _mm_loadu_si128((__m128i const *)(src + 1 * src_stride));
+  __m128i u2 = _mm_loadu_si128((__m128i const *)(src + 2 * src_stride));
+  __m128i u3 = _mm_loadu_si128((__m128i const *)(src + 3 * src_stride));
+  __m128i u4 = _mm_loadu_si128((__m128i const *)(src + 4 * src_stride));
+  __m128i u5 = _mm_loadu_si128((__m128i const *)(src + 5 * src_stride));
+  __m128i u6 = _mm_loadu_si128((__m128i const *)(src + 6 * src_stride));
+  __m128i u7 = _mm_loadu_si128((__m128i const *)(src + 7 * src_stride));
+
+  u0 = _mm_mulhrs_epi16(u0, k_256);
+  u1 = _mm_mulhrs_epi16(u1, k_256);
+  u2 = _mm_mulhrs_epi16(u2, k_256);
+  u3 = _mm_mulhrs_epi16(u3, k_256);
+  u4 = _mm_mulhrs_epi16(u4, k_256);
+  u5 = _mm_mulhrs_epi16(u5, k_256);
+  u6 = _mm_mulhrs_epi16(u6, k_256);
+  u7 = _mm_mulhrs_epi16(u7, k_256);
+
+  v0 = _mm_packus_epi16(u0, u1);
+  v1 = _mm_packus_epi16(u2, u3);
+  v2 = _mm_packus_epi16(u4, u5);
+  v3 = _mm_packus_epi16(u6, u7);
+
+  u0 = _mm_unpacklo_epi8(v0, v1);
+  u1 = _mm_unpackhi_epi8(v0, v1);
+  u2 = _mm_unpacklo_epi8(v2, v3);
+  u3 = _mm_unpackhi_epi8(v2, v3);
+
+  u4 = _mm_unpacklo_epi8(u0, u1);
+  u5 = _mm_unpacklo_epi8(u2, u3);
+  u6 = _mm_unpackhi_epi8(u0, u1);
+  u7 = _mm_unpackhi_epi8(u2, u3);
+
+  u0 = _mm_unpacklo_epi32(u4, u5);
+  u1 = _mm_unpackhi_epi32(u4, u5);
+  u2 = _mm_unpacklo_epi32(u6, u7);
+  u3 = _mm_unpackhi_epi32(u6, u7);
+
+  u4 = _mm_srli_si128(u0, 8);
+  u5 = _mm_srli_si128(u1, 8);
+  u6 = _mm_srli_si128(u2, 8);
+  u7 = _mm_srli_si128(u3, 8);
+
+  v0 = _mm_loadl_epi64((__m128i const *)(dst + 0 * dst_stride));
+  v1 = _mm_loadl_epi64((__m128i const *)(dst + 1 * dst_stride));
+  v2 = _mm_loadl_epi64((__m128i const *)(dst + 2 * dst_stride));
+  v3 = _mm_loadl_epi64((__m128i const *)(dst + 3 * dst_stride));
+  v4 = _mm_loadl_epi64((__m128i const *)(dst + 4 * dst_stride));
+  v5 = _mm_loadl_epi64((__m128i const *)(dst + 5 * dst_stride));
+  v6 = _mm_loadl_epi64((__m128i const *)(dst + 6 * dst_stride));
+  v7 = _mm_loadl_epi64((__m128i const *)(dst + 7 * dst_stride));
+
+  u0 = _mm_unpacklo_epi8(u0, zero);
+  u1 = _mm_unpacklo_epi8(u1, zero);
+  u2 = _mm_unpacklo_epi8(u2, zero);
+  u3 = _mm_unpacklo_epi8(u3, zero);
+  u4 = _mm_unpacklo_epi8(u4, zero);
+  u5 = _mm_unpacklo_epi8(u5, zero);
+  u6 = _mm_unpacklo_epi8(u6, zero);
+  u7 = _mm_unpacklo_epi8(u7, zero);
+
+  v0 = _mm_unpacklo_epi8(v0, zero);
+  v1 = _mm_unpacklo_epi8(v1, zero);
+  v2 = _mm_unpacklo_epi8(v2, zero);
+  v3 = _mm_unpacklo_epi8(v3, zero);
+  v4 = _mm_unpacklo_epi8(v4, zero);
+  v5 = _mm_unpacklo_epi8(v5, zero);
+  v6 = _mm_unpacklo_epi8(v6, zero);
+  v7 = _mm_unpacklo_epi8(v7, zero);
+
+  v0 = _mm_adds_epi16(u0, v0);
+  v1 = _mm_adds_epi16(u4, v1);
+  v2 = _mm_adds_epi16(u1, v2);
+  v3 = _mm_adds_epi16(u5, v3);
+  v4 = _mm_adds_epi16(u2, v4);
+  v5 = _mm_adds_epi16(u6, v5);
+  v6 = _mm_adds_epi16(u3, v6);
+  v7 = _mm_adds_epi16(u7, v7);
+
+  v0 = _mm_adds_epi16(v0, one);
+  v1 = _mm_adds_epi16(v1, one);
+  v2 = _mm_adds_epi16(v2, one);
+  v3 = _mm_adds_epi16(v3, one);
+  v4 = _mm_adds_epi16(v4, one);
+  v5 = _mm_adds_epi16(v5, one);
+  v6 = _mm_adds_epi16(v6, one);
+  v7 = _mm_adds_epi16(v7, one);
+
+  v0 = _mm_srai_epi16(v0, 1);
+  v1 = _mm_srai_epi16(v1, 1);
+  v2 = _mm_srai_epi16(v2, 1);
+  v3 = _mm_srai_epi16(v3, 1);
+  v4 = _mm_srai_epi16(v4, 1);
+  v5 = _mm_srai_epi16(v5, 1);
+  v6 = _mm_srai_epi16(v6, 1);
+  v7 = _mm_srai_epi16(v7, 1);
+
+  u0 = _mm_packus_epi16(v0, v1);
+  u1 = _mm_packus_epi16(v2, v3);
+  u2 = _mm_packus_epi16(v4, v5);
+  u3 = _mm_packus_epi16(v6, v7);
+
+  u4 = _mm_srli_si128(u0, 8);
+  u5 = _mm_srli_si128(u1, 8);
+  u6 = _mm_srli_si128(u2, 8);
+  u7 = _mm_srli_si128(u3, 8);
+
+  _mm_storel_epi64((__m128i*)dst, u0);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 1), u4);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 2), u1);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 3), u5);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 4), u2);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 5), u6);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 6), u3);
+  _mm_storel_epi64((__m128i*)(dst + dst_stride * 7), u7);
+}
+
+static transpose_to_dst_t trans8x8Tab[2] = {
+  transpose8x8_direct_to_dst, transpose8x8_accumu_to_dst
+};
+
+static INLINE void transpose_8x16(const __m128i *in, __m128i *out) {
+  __m128i t0, t1, t2, t3, u0, u1;
+
+  t0 = _mm_unpacklo_epi16(in[0], in[1]);
+  t1 = _mm_unpacklo_epi16(in[2], in[3]);
+  t2 = _mm_unpacklo_epi16(in[4], in[5]);
+  t3 = _mm_unpacklo_epi16(in[6], in[7]);
+
+  u0 = _mm_unpacklo_epi32(t0, t1);
+  u1 = _mm_unpacklo_epi32(t2, t3);
+
+  out[0] = _mm_unpacklo_epi64(u0, u1);
+  out[1] = _mm_unpackhi_epi64(u0, u1);
+
+  u0 = _mm_unpackhi_epi32(t0, t1);
+  u1 = _mm_unpackhi_epi32(t2, t3);
+
+  out[2] = _mm_unpacklo_epi64(u0, u1);
+  out[3] = _mm_unpackhi_epi64(u0, u1);
+
+  t0 = _mm_unpackhi_epi16(in[0], in[1]);
+  t1 = _mm_unpackhi_epi16(in[2], in[3]);
+  t2 = _mm_unpackhi_epi16(in[4], in[5]);
+  t3 = _mm_unpackhi_epi16(in[6], in[7]);
+
+  u0 = _mm_unpacklo_epi32(t0, t1);
+  u1 = _mm_unpacklo_epi32(t2, t3);
+
+  out[4] = _mm_unpacklo_epi64(u0, u1);
+  out[5] = _mm_unpackhi_epi64(u0, u1);
+
+  // Ignore out[6] and out[7]
+  // they're zero vectors.
+}
+
+static void filter_horiz_v8p_ssse3(const uint8_t *src_ptr, ptrdiff_t src_pitch,
+                                   __m128i *f, int tapsNum, uint16_t *buf) {
+  __m128i s[8], t[6];
+  __m128i min_x2x3, max_x2x3;
+  __m128i temp;
+
+  if (tapsNum == 10) {
+    src_ptr -= 1;
+  }
+  s[0] = _mm_loadu_si128((const __m128i *)src_ptr);
+  s[1] = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch));
+  s[2] = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2));
+  s[3] = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3));
+  s[4] = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 4));
+  s[5] = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 5));
+  s[6] = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 6));
+  s[7] = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 7));
+
+  // TRANSPOSE...
+  // Vecotor represents column pixel pairs instead of a row
+  transpose_8x16(s, t);
+
+  // multiply 2 adjacent elements with the filter and add the result
+  s[0] = _mm_maddubs_epi16(t[0], f[0]);
+  s[1] = _mm_maddubs_epi16(t[1], f[1]);
+  s[2] = _mm_maddubs_epi16(t[2], f[2]);
+  s[3] = _mm_maddubs_epi16(t[3], f[3]);
+  s[4] = _mm_maddubs_epi16(t[4], f[4]);
+  s[5] = _mm_maddubs_epi16(t[5], f[5]);
+
+  // add and saturate the results together
+  min_x2x3 = _mm_min_epi16(s[2], s[3]);
+  max_x2x3 = _mm_max_epi16(s[2], s[3]);
+  temp = _mm_adds_epi16(s[0], s[1]);
+  temp = _mm_adds_epi16(temp, s[5]);
+  temp = _mm_adds_epi16(temp, s[4]);
+
+  temp = _mm_adds_epi16(temp, min_x2x3);
+  temp = _mm_adds_epi16(temp, max_x2x3);
+
+  _mm_storeu_si128((__m128i *)buf, temp);
+}
+
+// Vertical 4-pixel parallel
+static INLINE void transpose4x4_direct_to_dst(const uint16_t *src,
+                                              int src_stride,
+                                              uint8_t *dst,
+                                              int dst_stride) {
+  const __m128i k_256 = _mm_set1_epi16(1 << 8);
+  __m128i v0, v1, v2, v3;
+
+  // TODO(luoyi): two loads, 8 elements per load (two bytes per element)
+  __m128i u0 = _mm_loadl_epi64((__m128i const *)(src + 0 * src_stride));
+  __m128i u1 = _mm_loadl_epi64((__m128i const *)(src + 1 * src_stride));
+  __m128i u2 = _mm_loadl_epi64((__m128i const *)(src + 2 * src_stride));
+  __m128i u3 = _mm_loadl_epi64((__m128i const *)(src + 3 * src_stride));
+
+  v0 = _mm_unpacklo_epi16(u0, u1);
+  v1 = _mm_unpacklo_epi16(u2, u3);
+
+  v2 = _mm_unpacklo_epi32(v0, v1);
+  v3 = _mm_unpackhi_epi32(v0, v1);
+
+  u0 = _mm_mulhrs_epi16(v2, k_256);
+  u1 = _mm_mulhrs_epi16(v3, k_256);
+
+  u0 = _mm_packus_epi16(u0, u1);
+  u1 = _mm_srli_si128(u0, 4);
+  u2 = _mm_srli_si128(u0, 8);
+  u3 = _mm_srli_si128(u0, 12);
+
+  *(int *)(dst) = _mm_cvtsi128_si32(u0);
+  *(int *)(dst + dst_stride) = _mm_cvtsi128_si32(u1);
+  *(int *)(dst + dst_stride * 2) = _mm_cvtsi128_si32(u2);
+  *(int *)(dst + dst_stride * 3) = _mm_cvtsi128_si32(u3);
+}
+
+static INLINE void transpose4x4_accumu_to_dst(const uint16_t *src,
+                                              int src_stride,
+                                              uint8_t *dst,
+                                              int dst_stride) {
+  const __m128i k_256 = _mm_set1_epi16(1 << 8);
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i one = _mm_set1_epi16(1);
+
+  __m128i v0, v1, v2, v3;
+
+  __m128i u0 = _mm_loadl_epi64((__m128i const *)(src));
+  __m128i u1 = _mm_loadl_epi64((__m128i const *)(src + src_stride));
+  __m128i u2 = _mm_loadl_epi64((__m128i const *)(src + 2 * src_stride));
+  __m128i u3 = _mm_loadl_epi64((__m128i const *)(src + 3 * src_stride));
+
+  v0 = _mm_unpacklo_epi16(u0, u1);
+  v1 = _mm_unpacklo_epi16(u2, u3);
+
+  v2 = _mm_unpacklo_epi32(v0, v1);
+  v3 = _mm_unpackhi_epi32(v0, v1);
+
+  u0 = _mm_mulhrs_epi16(v2, k_256);
+  u1 = _mm_mulhrs_epi16(v3, k_256);
+
+  u2 = _mm_packus_epi16(u0, u1);
+  u0 = _mm_unpacklo_epi8(u2, zero);
+  u1 = _mm_unpackhi_epi8(u2, zero);
+
+  // load pixel values
+  v0 = _mm_loadl_epi64((__m128i const *)(dst));
+  v1 = _mm_loadl_epi64((__m128i const *)(dst + dst_stride));
+  v2 = _mm_loadl_epi64((__m128i const *)(dst + 2 * dst_stride));
+  v3 = _mm_loadl_epi64((__m128i const *)(dst + 3 * dst_stride));
+
+  v0 = _mm_unpacklo_epi8(v0, zero);
+  v1 = _mm_unpacklo_epi8(v1, zero);
+  v2 = _mm_unpacklo_epi8(v2, zero);
+  v3 = _mm_unpacklo_epi8(v3, zero);
+
+  v0 = _mm_unpacklo_epi64(v0, v1);
+  v1 = _mm_unpacklo_epi64(v2, v3);
+
+  u0 = _mm_adds_epi16(u0, v0);
+  u1 = _mm_adds_epi16(u1, v1);
+
+  u0 = _mm_adds_epi16(u0, one);
+  u1 = _mm_adds_epi16(u1, one);
+
+  u0 = _mm_srai_epi16(u0, 1);
+  u1 = _mm_srai_epi16(u1, 1);
+
+  // saturation and pack to pixels
+  u0 = _mm_packus_epi16(u0, u1);
+  u1 = _mm_srli_si128(u0, 4);
+  u2 = _mm_srli_si128(u0, 8);
+  u3 = _mm_srli_si128(u0, 12);
+
+  *(int *)(dst) = _mm_cvtsi128_si32(u0);
+  *(int *)(dst + dst_stride) = _mm_cvtsi128_si32(u1);
+  *(int *)(dst + dst_stride * 2) = _mm_cvtsi128_si32(u2);
+  *(int *)(dst + dst_stride * 3) = _mm_cvtsi128_si32(u3);
+}
+
+static transpose_to_dst_t trans4x4Tab[2] = {
+  transpose4x4_direct_to_dst, transpose4x4_accumu_to_dst
+};
+
+static void filter_horiz_v4p_ssse3(const uint8_t *src_ptr, ptrdiff_t src_pitch,
+                                   __m128i *f, int tapsNum, uint16_t *buf) {
+  __m128i A, B, C, D;
+  __m128i tr0_0, tr0_1, s1s0, s3s2, s5s4, s7s6, s9s8, sbsa;
+  __m128i x0, x1, x2, x3, x4, x5;
+  __m128i min_x2x3, max_x2x3, temp;
+
+  if (tapsNum == 10) {
+    src_ptr -= 1;
+  }
+  A = _mm_loadu_si128((const __m128i *)src_ptr);
+  B = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch));
+  C = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2));
+  D = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3));
+
+  // TRANSPOSE...
+  // Vecotor represents column pixel pairs instead of a row
+  // 00 01 10 11 02 03 12 13 04 05 14 15 06 07 16 17
+  tr0_0 = _mm_unpacklo_epi16(A, B);
+  // 20 21 30 31 22 23 32 33 24 25 34 35 26 27 36 37
+  tr0_1 = _mm_unpacklo_epi16(C, D);
+  // 00 01 10 11 20 21 30 31 02 03 12 13 22 23 32 33
+  s1s0  = _mm_unpacklo_epi32(tr0_0, tr0_1);
+  // 04 05 14 15 24 25 34 35 06 07 16 17 26 27 36 37
+  s5s4 = _mm_unpackhi_epi32(tr0_0, tr0_1);
+  // 02 03 12 13 22 23 32 33
+  s3s2 = _mm_srli_si128(s1s0, 8);
+  // 06 07 16 17 26 27 36 37
+  s7s6 = _mm_srli_si128(s5s4, 8);
+
+  tr0_0 = _mm_unpackhi_epi16(A, B);
+  tr0_1 = _mm_unpackhi_epi16(C, D);
+  s9s8  = _mm_unpacklo_epi32(tr0_0, tr0_1);
+  sbsa = _mm_srli_si128(s9s8, 8);
+
+  // multiply 2 adjacent elements with the filter and add the result
+  x0 = _mm_maddubs_epi16(s1s0, f[0]);
+  x1 = _mm_maddubs_epi16(s3s2, f[1]);
+  x2 = _mm_maddubs_epi16(s5s4, f[2]);
+  x3 = _mm_maddubs_epi16(s7s6, f[3]);
+  x4 = _mm_maddubs_epi16(s9s8, f[4]);
+  x5 = _mm_maddubs_epi16(sbsa, f[5]);
+  // add and saturate the results together
+  min_x2x3 = _mm_min_epi16(x2, x3);
+  max_x2x3 = _mm_max_epi16(x2, x3);
+  temp = _mm_adds_epi16(x0, x1);
+  temp = _mm_adds_epi16(temp, x5);
+  temp = _mm_adds_epi16(temp, x4);
+
+  temp = _mm_adds_epi16(temp, min_x2x3);
+  temp = _mm_adds_epi16(temp, max_x2x3);
+  _mm_storel_epi64((__m128i *)buf, temp);
+}
+
+// Note:
+//  This function assumes:
+// (1) 10/12-taps filters
+// (2) x_step_q4 = 16 then filter is fixed at the call
+
+void vp10_convolve_horiz_ssse3(const uint8_t *src, int src_stride, uint8_t *dst,
+                               int dst_stride, int w, int h,
+                               const InterpFilterParams filter_params,
+                               const int subpel_x_q4, int x_step_q4, int avg) {
+  DECLARE_ALIGNED(16, uint16_t, temp[8 * 8]);
+  __m128i verf[6];
+  __m128i horf[2];
+  SubpelFilterCoeffs hCoeffs, vCoeffs;
+  const uint8_t *src_ptr;
+  store_pixel_t store4p = store4pixelTab[avg];
+  transpose_to_dst_t transpose_4x4 = trans4x4Tab[avg];
+  transpose_to_dst_t transpose_8x8 = trans8x8Tab[avg];
+
+  const int tapsNum = filter_params.taps;
+  int block_height, block_residu;
+  int i, col, count;
+  (void)x_step_q4;
+
+  if (0 == subpel_x_q4 || 16 != x_step_q4) {
+    vp10_convolve_horiz_c(src, src_stride, dst, dst_stride, w, h, filter_params,
+                          subpel_x_q4, x_step_q4, avg);
+    return;
+  }
+
+  hCoeffs = vp10_get_subpel_filter_signal_dir(
+      filter_params, subpel_x_q4 - 1);
+  vCoeffs = vp10_get_subpel_filter_ver_signal_dir(
+      filter_params, subpel_x_q4 - 1);
+
+  if (!hCoeffs || !vCoeffs) {
+    vp10_convolve_horiz_c(src, src_stride, dst, dst_stride, w, h, filter_params,
+                          subpel_x_q4, x_step_q4, avg);
+    return;
+  }
+
+  verf[0] = *((const __m128i *)(vCoeffs));
+  verf[1] = *((const __m128i *)(vCoeffs + 1));
+  verf[2] = *((const __m128i *)(vCoeffs + 2));
+  verf[3] = *((const __m128i *)(vCoeffs + 3));
+  verf[4] = *((const __m128i *)(vCoeffs + 4));
+  verf[5] = *((const __m128i *)(vCoeffs + 5));
+
+  horf[0] = *((const __m128i *)(hCoeffs));
+  horf[1] = *((const __m128i *)(hCoeffs + 1));
+
+  count = 0;
+
+  // here tapsNum is filter size
+  src -= (tapsNum >> 1) - 1;
+  src_ptr = src;
+  if (w > WIDTH_BOUND && h > HEIGHT_BOUND) {
+    // 8-pixels parallel
+    block_height = h >> 3;
+    block_residu = h & 7;
+
+    do {
+      for (col = 0; col < w; col += 8) {
+        for (i = 0; i < 8; ++i) {
+          filter_horiz_v8p_ssse3(src_ptr, src_stride, verf, tapsNum,
+                                 temp + (i * 8));
+          src_ptr += 1;
+        }
+        transpose_8x8(temp, 8, dst + col, dst_stride);
+      }
+      count++;
+      src_ptr = src + count * src_stride * 8;
+      dst += dst_stride * 8;
+    } while (count < block_height);
+
+    for (i = 0; i < block_residu; ++i) {
+      filter_horiz_ssse3(src_ptr, horf, tapsNum, w, store4p, dst, dst);
+      src_ptr += src_stride;
+      dst += dst_stride;
+    }
+  } else {
+    // 4-pixels parallel
+    block_height = h >> 2;
+    block_residu = h & 3;
+
+    do {
+      for (col = 0; col < w; col += 4) {
+        for (i = 0; i < 4; ++i) {
+          filter_horiz_v4p_ssse3(src_ptr, src_stride, verf, tapsNum,
+                                 temp + (i * 4));
+          src_ptr += 1;
+        }
+        transpose_4x4(temp, 4, dst + col, dst_stride);
+      }
+      count++;
+      src_ptr = src + count * src_stride * 4;
+      dst += dst_stride * 4;
+    } while (count < block_height);
+
+    for (i = 0; i < block_residu; ++i) {
+      filter_horiz_ssse3(src_ptr, horf, tapsNum, w, store4p, dst, dst);
+      src_ptr += src_stride;
+      dst += dst_stride;
+    }
+  }
+}

diff --git a/vp10/vp10_common.mk b/vp10/vp10_common.mk
index 9bc0ce074264093000aa1d86e955a021d3d1fcde..0253b4ccf4d16de252333524020a44675bd4e87d 100644 (file)
--- a/vp10/vp10_common.mk
+++ b/vp10/vp10_common.mk
@@ -72,6 +72,8 @@ VP10_COMMON_SRCS-yes += common/vp10_fwd_txfm2d.c
 VP10_COMMON_SRCS-yes += common/vp10_fwd_txfm2d_cfg.h
 VP10_COMMON_SRCS-yes += common/vp10_inv_txfm2d.c
 VP10_COMMON_SRCS-yes += common/vp10_inv_txfm2d_cfg.h
+VP10_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp10_convolve_ssse3.c
+VP10_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp10_convolve_filters_ssse3.c
 VP10_COMMON_SRCS-yes += common/vp10_convolve.c
 VP10_COMMON_SRCS-yes += common/vp10_convolve.h
 VP10_COMMON_SRCS-$(CONFIG_ANS) += common/ans.h