SSSE3 convolution optimization

Optimizing all SSSE3 assembly for convolution:
1. vp9_filter_block1d4_h8_sse2
2. vp9_filter_block1d8_h8_sse2
3. vp9_filter_block1d16_h8_sse2
4. vp9_filter_block1d4_v8_sse2
5. vp9_filter_block1d8_v8_sse2
6. vp9_filter_block1d16_v8_sse2
my optimization include:
-processing 2x8 elements in one 128 bit register instead of processing
8 elements in one 128 bit register.
-removing unecessary loads.
This optimization gives between 2.4% user level gain for 480p input
and 1.6% user level gain for 720p.
This Optimization done only for 64bit.

Change-Id: Icb586dc0c938b56699864fcee6c52fd43b36b969

3 files changed, 685 insertions, 7 deletions

diff --git a/vp9/common/x86/vp9_asm_stubs.c b/vp9/common/x86/vp9_asm_stubs.c
index 106e6d426..b4ab4c306 100644
--- a/vp9/common/x86/vp9_asm_stubs.c
+++ b/vp9/common/x86/vp9_asm_stubs.c
@@ -45,20 +45,105 @@ typedef void filter8_1dfunction (
   const short *filter
 );
 
-#if HAVE_SSSE3
-filter8_1dfunction vp9_filter_block1d16_v8_ssse3;
-filter8_1dfunction vp9_filter_block1d16_h8_ssse3;
-filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
-filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
-filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
-filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
+#if (HAVE_SSSE3)
 filter8_1dfunction vp9_filter_block1d16_v8_avg_ssse3;
 filter8_1dfunction vp9_filter_block1d16_h8_avg_ssse3;
 filter8_1dfunction vp9_filter_block1d8_v8_avg_ssse3;
 filter8_1dfunction vp9_filter_block1d8_h8_avg_ssse3;
 filter8_1dfunction vp9_filter_block1d4_v8_avg_ssse3;
 filter8_1dfunction vp9_filter_block1d4_h8_avg_ssse3;
+#if (ARCH_X86_64)
+filter8_1dfunction vp9_filter_block1d16_v8_intrin_ssse3;
+filter8_1dfunction vp9_filter_block1d16_h8_intrin_ssse3;
+filter8_1dfunction vp9_filter_block1d8_v8_intrin_ssse3;
+filter8_1dfunction vp9_filter_block1d8_h8_intrin_ssse3;
+filter8_1dfunction vp9_filter_block1d4_v8_intrin_ssse3;
+filter8_1dfunction vp9_filter_block1d4_h8_intrin_ssse3;
+
+void vp9_convolve8_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
+                               uint8_t *dst, ptrdiff_t dst_stride,
+                               const int16_t *filter_x, int x_step_q4,
+                               const int16_t *filter_y, int y_step_q4,
+                               int w, int h) {
+  /* Ensure the filter can be compressed to int16_t. */
+  if (x_step_q4 == 16 && filter_x[3] != 128) {
+    while (w >= 16) {
+      vp9_filter_block1d16_h8_intrin_ssse3(src, src_stride,
+                                    dst, dst_stride,
+                                    h, filter_x);
+      src += 16;
+      dst += 16;
+      w -= 16;
+    }
+    while (w >= 8) {
+      vp9_filter_block1d8_h8_intrin_ssse3(src, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_x);
+      src += 8;
+      dst += 8;
+      w -= 8;
+    }
+    while (w >= 4) {
+      vp9_filter_block1d4_h8_intrin_ssse3(src, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_x);
+      src += 4;
+      dst += 4;
+      w -= 4;
+    }
+  }
+  if (w) {
+    vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride,
+                          filter_x, x_step_q4, filter_y, y_step_q4,
+                          w, h);
+  }
+}
 
+void vp9_convolve8_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
+                              uint8_t *dst, ptrdiff_t dst_stride,
+                              const int16_t *filter_x, int x_step_q4,
+                              const int16_t *filter_y, int y_step_q4,
+                              int w, int h) {
+  if (y_step_q4 == 16 && filter_y[3] != 128) {
+    while (w >= 16) {
+      vp9_filter_block1d16_v8_intrin_ssse3(src - src_stride * 3, src_stride,
+                                    dst, dst_stride,
+                                    h, filter_y);
+      src += 16;
+      dst += 16;
+      w -= 16;
+    }
+    while (w >= 8) {
+      vp9_filter_block1d8_v8_intrin_ssse3(src - src_stride * 3, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_y);
+      src += 8;
+      dst += 8;
+      w -= 8;
+    }
+    while (w >= 4) {
+      vp9_filter_block1d4_v8_intrin_ssse3(src - src_stride * 3, src_stride,
+                                   dst, dst_stride,
+                                   h, filter_y);
+      src += 4;
+      dst += 4;
+      w -= 4;
+    }
+  }
+  if (w) {
+    vp9_convolve8_vert_c(src, src_stride, dst, dst_stride,
+                         filter_x, x_step_q4, filter_y, y_step_q4,
+                         w, h);
+  }
+}
+
+#else
+filter8_1dfunction vp9_filter_block1d16_v8_ssse3;
+filter8_1dfunction vp9_filter_block1d16_h8_ssse3;
+filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
+filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
+filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
+filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
 void vp9_convolve8_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
                                uint8_t *dst, ptrdiff_t dst_stride,
                                const int16_t *filter_x, int x_step_q4,
@@ -135,6 +220,7 @@ void vp9_convolve8_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
                          w, h);
   }
 }
+#endif
 
 void vp9_convolve8_avg_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
                                uint8_t *dst, ptrdiff_t dst_stride,
diff --git a/vp9/common/x86/vp9_subpixel_8t_intrin_ssse3.c b/vp9/common/x86/vp9_subpixel_8t_intrin_ssse3.c
new file mode 100644
index 000000000..303fced3b
--- /dev/null
+++ b/vp9/common/x86/vp9_subpixel_8t_intrin_ssse3.c
@@ -0,0 +1,591 @@
+/*
+ *  Copyright (c) 2010 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 <tmmintrin.h>
+#include "vpx_ports/mem.h"
+#include "vpx_ports/emmintrin_compat.h"
+
+
+// filters only for the 4_h8 convolution
+DECLARE_ALIGNED(16, const unsigned char,
+filt1_4_h8[16])= {0, 1, 1, 2, 2, 3, 3, 4, 2, 3, 3, 4, 4, 5, 5, 6};
+
+DECLARE_ALIGNED(16, const unsigned char,
+filt2_4_h8[16])= {4, 5, 5, 6, 6, 7, 7, 8, 6, 7, 7, 8, 8, 9, 9, 10};
+
+// filters for 8_h8 and 16_h8
+DECLARE_ALIGNED(16, const unsigned char,
+filt1_global[16])= {0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8};
+
+DECLARE_ALIGNED(16, const unsigned char,
+filt2_global[16])= {2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10};
+
+DECLARE_ALIGNED(16, const unsigned char,
+filt3_global[16])= {4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12};
+
+DECLARE_ALIGNED(16, const unsigned char,
+filt4_global[16])= {6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14};
+
+
+
+void vp9_filter_block1d4_h8_intrin_ssse3(unsigned char *src_ptr,
+                                         unsigned int src_pixels_per_line,
+                                         unsigned char *output_ptr,
+                                         unsigned int output_pitch,
+                                         unsigned int output_height,
+                                         int16_t *filter) {
+    __m128i firstFilters, secondFilters, thirdFilters, forthFilters;
+    __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt4;
+    __m128i addFilterReg64, filtersReg, srcReg, minReg;
+    unsigned int i;
+
+    // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+    addFilterReg64 =_mm_set1_epi32((int)0x0400040u);
+    filtersReg = _mm_loadu_si128((__m128i *)filter);
+    // converting the 16 bit (short) to  8 bit (byte) and have the same data
+    // in both lanes of 128 bit register.
+    filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+
+    // duplicate only the first 16 bits in the filter into the first lane
+    firstFilters = _mm_shufflelo_epi16(filtersReg, 0);
+    // duplicate only the third 16 bit in the filter into the first lane
+    secondFilters = _mm_shufflelo_epi16(filtersReg, 0xAAu);
+    // duplicate only the seconds 16 bits in the filter into the second lane
+    firstFilters = _mm_shufflehi_epi16(firstFilters, 0x55u);
+    // duplicate only the forth 16 bits in the filter into the second lane
+    secondFilters = _mm_shufflehi_epi16(secondFilters, 0xFFu);
+
+    // loading the local filters
+    thirdFilters =_mm_load_si128((__m128i const *)filt1_4_h8);
+    forthFilters = _mm_load_si128((__m128i const *)filt2_4_h8);
+
+    for (i = 0; i < output_height; i++) {
+        srcReg = _mm_loadu_si128((__m128i *)(src_ptr-3));
+
+        // filter the source buffer
+        srcRegFilt1= _mm_shuffle_epi8(srcReg, thirdFilters);
+        srcRegFilt2= _mm_shuffle_epi8(srcReg, forthFilters);
+
+        // multiply 2 adjacent elements with the filter and add the result
+        srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
+        srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, secondFilters);
+
+        // extract the higher half of the lane
+        srcRegFilt3 =  _mm_srli_si128(srcRegFilt1, 8);
+        srcRegFilt4 =  _mm_srli_si128(srcRegFilt2, 8);
+
+        minReg = _mm_min_epi16(srcRegFilt3, srcRegFilt2);
+
+        // add and saturate all the results together
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt4);
+
+        srcRegFilt3 = _mm_max_epi16(srcRegFilt3, srcRegFilt2);
+
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg);
+
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt3);
+
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
+
+        // shift by 7 bit each 16 bits
+        srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
+
+        // shrink to 8 bit each 16 bits
+        srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
+
+        src_ptr+=src_pixels_per_line;
+
+        // save only 4 bytes
+        *((int*)&output_ptr[0])= _mm_cvtsi128_si32(srcRegFilt1);
+
+        output_ptr+=output_pitch;
+    }
+}
+
+
+void vp9_filter_block1d8_h8_intrin_ssse3(unsigned char *src_ptr,
+                                         unsigned int src_pixels_per_line,
+                                         unsigned char *output_ptr,
+                                         unsigned int output_pitch,
+                                         unsigned int output_height,
+                                         int16_t *filter) {
+    __m128i firstFilters, secondFilters, thirdFilters, forthFilters, srcReg;
+    __m128i filt1Reg, filt2Reg, filt3Reg, filt4Reg;
+    __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt4;
+    __m128i addFilterReg64, filtersReg, minReg;
+    unsigned int i;
+
+    // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+    addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
+    filtersReg = _mm_loadu_si128((__m128i *)filter);
+    // converting the 16 bit (short) to  8 bit (byte) and have the same data
+    // in both lanes of 128 bit register.
+    filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+
+    // duplicate only the first 16 bits (first and second byte)
+    // across 128 bit register
+    firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u));
+    // duplicate only the second 16 bits (third and forth byte)
+    // across 128 bit register
+    secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
+    // duplicate only the third 16 bits (fifth and sixth byte)
+    // across 128 bit register
+    thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
+    // duplicate only the forth 16 bits (seventh and eighth byte)
+    // across 128 bit register
+    forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
+
+    filt1Reg = _mm_load_si128((__m128i const *)filt1_global);
+    filt2Reg = _mm_load_si128((__m128i const *)filt2_global);
+    filt3Reg = _mm_load_si128((__m128i const *)filt3_global);
+    filt4Reg = _mm_load_si128((__m128i const *)filt4_global);
+
+    for (i = 0; i < output_height; i++) {
+        srcReg = _mm_loadu_si128((__m128i *)(src_ptr-3));
+
+        // filter the source buffer
+        srcRegFilt1= _mm_shuffle_epi8(srcReg, filt1Reg);
+        srcRegFilt2= _mm_shuffle_epi8(srcReg, filt2Reg);
+
+        // multiply 2 adjacent elements with the filter and add the result
+        srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
+        srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, secondFilters);
+
+        // filter the source buffer
+        srcRegFilt3= _mm_shuffle_epi8(srcReg, filt3Reg);
+        srcRegFilt4= _mm_shuffle_epi8(srcReg, filt4Reg);
+
+        // multiply 2 adjacent elements with the filter and add the result
+        srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, thirdFilters);
+        srcRegFilt4 = _mm_maddubs_epi16(srcRegFilt4, forthFilters);
+
+        // add and saturate all the results together
+        minReg = _mm_min_epi16(srcRegFilt4, srcRegFilt3);
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt2);
+
+        srcRegFilt4= _mm_max_epi16(srcRegFilt4, srcRegFilt3);
+
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg);
+
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt4);
+
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
+
+        // shift by 7 bit each 16 bits
+        srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
+
+        // shrink to 8 bit each 16 bits
+        srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
+
+        src_ptr+=src_pixels_per_line;
+
+       // save only 8 bytes
+        _mm_storel_epi64((__m128i*)&output_ptr[0], srcRegFilt1);
+
+        output_ptr+=output_pitch;
+    }
+}
+
+void vp9_filter_block1d16_h8_intrin_ssse3(unsigned char *src_ptr,
+                                          unsigned int src_pixels_per_line,
+                                          unsigned char *output_ptr,
+                                          unsigned int output_pitch,
+                                          unsigned int output_height,
+                                          int16_t *filter) {
+    __m128i addFilterReg64, filtersReg, srcReg1, srcReg2;
+    __m128i filt1Reg, filt2Reg, filt3Reg, filt4Reg;
+    __m128i firstFilters, secondFilters, thirdFilters, forthFilters;
+    __m128i srcRegFilt1_1, srcRegFilt2_1, srcRegFilt2, srcRegFilt3;
+    unsigned int i;
+
+    // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+    addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
+    filtersReg = _mm_loadu_si128((__m128i *)filter);
+    // converting the 16 bit (short) to  8 bit (byte) and have the same data
+    // in both lanes of 128 bit register.
+    filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+
+    // duplicate only the first 16 bits (first and second byte)
+    // across 128 bit register
+    firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u));
+    // duplicate only the second 16 bits (third and forth byte)
+    // across 128 bit register
+    secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
+    // duplicate only the third 16 bits (fifth and sixth byte)
+    // across 128 bit register
+    thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
+    // duplicate only the forth 16 bits (seventh and eighth byte)
+    // across 128 bit register
+    forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
+
+    filt1Reg = _mm_load_si128((__m128i const *)filt1_global);
+    filt2Reg = _mm_load_si128((__m128i const *)filt2_global);
+    filt3Reg = _mm_load_si128((__m128i const *)filt3_global);
+    filt4Reg = _mm_load_si128((__m128i const *)filt4_global);
+
+    for (i = 0; i < output_height; i++) {
+        srcReg1 = _mm_loadu_si128((__m128i *)(src_ptr-3));
+
+        // filter the source buffer
+        srcRegFilt1_1= _mm_shuffle_epi8(srcReg1, filt1Reg);
+        srcRegFilt2= _mm_shuffle_epi8(srcReg1, filt2Reg);
+
+        // multiply 2 adjacent elements with the filter and add the result
+        srcRegFilt1_1 = _mm_maddubs_epi16(srcRegFilt1_1, firstFilters);
+        srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, secondFilters);
+
+        // add and saturate the results together
+        srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, srcRegFilt2);
+
+        // filter the source buffer
+        srcRegFilt3= _mm_shuffle_epi8(srcReg1, filt4Reg);
+        srcRegFilt2= _mm_shuffle_epi8(srcReg1, filt3Reg);
+
+        // multiply 2 adjacent elements with the filter and add the result
+        srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, forthFilters);
+        srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, thirdFilters);
+
+        // add and saturate the results together
+        srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
+        _mm_min_epi16(srcRegFilt3, srcRegFilt2));
+
+        // reading the next 16 bytes.
+        // (part of it was being read by earlier read)
+        srcReg2 = _mm_loadu_si128((__m128i *)(src_ptr+5));
+
+        // add and saturate the results together
+        srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
+        _mm_max_epi16(srcRegFilt3, srcRegFilt2));
+
+        // filter the source buffer
+        srcRegFilt2_1= _mm_shuffle_epi8(srcReg2, filt1Reg);
+        srcRegFilt2= _mm_shuffle_epi8(srcReg2, filt2Reg);
+
+        // multiply 2 adjacent elements with the filter and add the result
+        srcRegFilt2_1 = _mm_maddubs_epi16(srcRegFilt2_1, firstFilters);
+        srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, secondFilters);
+
+        // add and saturate the results together
+        srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, srcRegFilt2);
+
+        // filter the source buffer
+        srcRegFilt3= _mm_shuffle_epi8(srcReg2, filt4Reg);
+        srcRegFilt2= _mm_shuffle_epi8(srcReg2, filt3Reg);
+
+        // multiply 2 adjacent elements with the filter and add the result
+        srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, forthFilters);
+        srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, thirdFilters);
+
+        // add and saturate the results together
+        srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
+        _mm_min_epi16(srcRegFilt3, srcRegFilt2));
+        srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
+        _mm_max_epi16(srcRegFilt3, srcRegFilt2));
+
+        srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, addFilterReg64);
+
+        srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, addFilterReg64);
+
+        // shift by 7 bit each 16 bit
+        srcRegFilt1_1 = _mm_srai_epi16(srcRegFilt1_1, 7);
+        srcRegFilt2_1 = _mm_srai_epi16(srcRegFilt2_1, 7);
+
+        // shrink to 8 bit each 16 bits, the first lane contain the first
+        // convolve result and the second lane contain the second convolve
+        // result
+        srcRegFilt1_1 = _mm_packus_epi16(srcRegFilt1_1, srcRegFilt2_1);
+
+        src_ptr+=src_pixels_per_line;
+
+        // save 16 bytes
+        _mm_store_si128((__m128i*)output_ptr, srcRegFilt1_1);
+
+        output_ptr+=output_pitch;
+    }
+}
+
+
+
+void vp9_filter_block1d4_v8_intrin_ssse3(unsigned char *src_ptr,
+                                         unsigned int src_pitch,
+                                         unsigned char *output_ptr,
+                                         unsigned int out_pitch,
+                                         unsigned int output_height,
+                                         int16_t *filter) {
+    __m128i addFilterReg64, filtersReg, firstFilters, secondFilters;
+    __m128i minReg, srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt4;
+    unsigned int i;
+
+    // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+    addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
+    filtersReg = _mm_loadu_si128((__m128i *)filter);
+    // converting the 16 bit (short) to  8 bit (byte) and have the same data
+    // in both lanes of 128 bit register.
+    filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+
+    // duplicate only the first 16 bits in the filter into the first lane
+    firstFilters = _mm_shufflelo_epi16(filtersReg, 0);
+    // duplicate only the second 16 bits in the filter into the second lane
+    firstFilters = _mm_shufflehi_epi16(firstFilters, 0x55u);
+    // duplicate only the third 16 bits in the filter into the first lane
+    secondFilters = _mm_shufflelo_epi16(filtersReg, 0xAAu);
+    // duplicate only the forth 16 bits in the filter into the second lane
+    secondFilters = _mm_shufflehi_epi16(secondFilters, 0xFFu);
+
+    for (i = 0; i < output_height; i++) {
+        // load the first 4 byte
+        srcRegFilt1 = _mm_cvtsi32_si128(*((int*)&src_ptr[0]));
+        // load the next 4 bytes in stride of src_pitch
+        srcRegFilt2 = _mm_cvtsi32_si128(*((int*)&(src_ptr+src_pitch)[0]));
+
+        // merge the result together
+        srcRegFilt1 = _mm_unpacklo_epi8(srcRegFilt1, srcRegFilt2);
+
+
+        srcRegFilt2 = _mm_cvtsi32_si128(*((int*)&(src_ptr+src_pitch*2)[0]));
+        srcRegFilt3 =  _mm_cvtsi32_si128(*((int*)&(src_ptr+src_pitch*3)[0]));
+
+        // merge the result together
+        srcRegFilt2 = _mm_unpacklo_epi8(srcRegFilt2, srcRegFilt3);
+
+        srcRegFilt3 = _mm_cvtsi32_si128(*((int*)&(src_ptr+src_pitch*4)[0]));
+        srcRegFilt4 = _mm_cvtsi32_si128(*((int*)&(src_ptr+src_pitch*5)[0]));
+
+        // merge the result together
+        srcRegFilt3 = _mm_unpacklo_epi8(srcRegFilt3, srcRegFilt4);
+        srcRegFilt1 = _mm_unpacklo_epi64(srcRegFilt1, srcRegFilt2);
+
+        srcRegFilt4 = _mm_cvtsi32_si128(*((int*)&(src_ptr+src_pitch*6)[0]));
+        srcRegFilt2 = _mm_cvtsi32_si128(*((int*)&(src_ptr+src_pitch*7)[0]));
+
+        // merge the result together
+        srcRegFilt4 = _mm_unpacklo_epi8(srcRegFilt4, srcRegFilt2);
+        srcRegFilt3 = _mm_unpacklo_epi64(srcRegFilt3, srcRegFilt4);
+
+        // multiply 2 adjacent elements with the filter and add the result
+        srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
+        srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, secondFilters);
+
+        // extract the second lane of the 128 bit register
+        srcRegFilt2 = _mm_srli_si128(srcRegFilt1, 8);
+
+        // add and saturate the results together
+        minReg = _mm_min_epi16(srcRegFilt2, srcRegFilt3);
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
+        _mm_srli_si128(srcRegFilt3, 8));
+        srcRegFilt2 = _mm_max_epi16(srcRegFilt2, srcRegFilt3);
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg);
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt2);
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
+
+        // shift by 7 bit each 16 bit
+        srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
+
+        // shrink to 8 bit each 16 bits
+        srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
+
+        src_ptr+=src_pitch;
+
+        // save only 4 bytes convolve result
+        *((int*)&output_ptr[0])= _mm_cvtsi128_si32(srcRegFilt1);
+
+        output_ptr+=out_pitch;
+    }
+}
+
+void vp9_filter_block1d8_v8_intrin_ssse3(unsigned char *src_ptr,
+                                         unsigned int src_pitch,
+                                         unsigned char *output_ptr,
+                                         unsigned int out_pitch,
+                                         unsigned int output_height,
+                                         int16_t *filter) {
+    __m128i addFilterReg64, filtersReg, minReg, srcRegFilt6;
+    __m128i firstFilters, secondFilters, thirdFilters, forthFilters;
+    __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt4, srcRegFilt5;
+    unsigned int i;
+
+    // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+    addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
+    filtersReg = _mm_loadu_si128((__m128i *)filter);
+    // converting the 16 bit (short) to  8 bit (byte) and have the same data
+    // in both lanes of 128 bit register.
+    filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+
+    // duplicate only the first 16 bits in the filter
+    firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u));
+    // duplicate only the second 16 bits in the filter
+    secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
+    // duplicate only the third 16 bits in the filter
+    thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
+    // duplicate only the forth 16 bits in the filter
+    forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
+
+    for (i = 0; i < output_height; i++) {
+        // load the first 8 bytes
+        srcRegFilt1 = _mm_loadl_epi64((__m128i *)&src_ptr[0]);
+        // load the next 8 bytes in stride of src_pitch
+        srcRegFilt2 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch)[0]);
+        srcRegFilt3 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch*2)[0]);
+        srcRegFilt4 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch*3)[0]);
+
+        // merge the result together
+        srcRegFilt1 = _mm_unpacklo_epi8(srcRegFilt1, srcRegFilt2);
+        srcRegFilt3 = _mm_unpacklo_epi8(srcRegFilt3, srcRegFilt4);
+
+        // load the next 8 bytes in stride of src_pitch
+        srcRegFilt2 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch*4)[0]);
+        srcRegFilt4 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch*5)[0]);
+        srcRegFilt5 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch*6)[0]);
+        srcRegFilt6 = _mm_loadl_epi64((__m128i *)&(src_ptr+src_pitch*7)[0]);
+
+        // merge the result together
+        srcRegFilt2 = _mm_unpacklo_epi8(srcRegFilt2, srcRegFilt4);
+        srcRegFilt5 = _mm_unpacklo_epi8(srcRegFilt5, srcRegFilt6);
+
+        // multiply 2 adjacent elements with the filter and add the result
+        srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
+        srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, secondFilters);
+        srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, thirdFilters);
+        srcRegFilt5 = _mm_maddubs_epi16(srcRegFilt5, forthFilters);
+
+        // add and saturate the results together
+        minReg = _mm_min_epi16(srcRegFilt2, srcRegFilt3);
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt5);
+        srcRegFilt2 = _mm_max_epi16(srcRegFilt2, srcRegFilt3);
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg);
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt2);
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
+
+        // shift by 7 bit each 16 bit
+        srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
+
+        // shrink to 8 bit each 16 bits
+        srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
+
+        src_ptr+=src_pitch;
+
+        // save only 8 bytes convolve result
+        _mm_storel_epi64((__m128i*)&output_ptr[0], srcRegFilt1);
+
+        output_ptr+=out_pitch;
+    }
+}
+
+
+void vp9_filter_block1d16_v8_intrin_ssse3(unsigned char *src_ptr,
+                                          unsigned int src_pitch,
+                                          unsigned char *output_ptr,
+                                          unsigned int out_pitch,
+                                          unsigned int output_height,
+                                          int16_t *filter) {
+    __m128i addFilterReg64, filtersReg, srcRegFilt1, srcRegFilt2, srcRegFilt3;
+    __m128i firstFilters, secondFilters, thirdFilters, forthFilters;
+    __m128i srcRegFilt4, srcRegFilt5, srcRegFilt6, srcRegFilt7, srcRegFilt8;
+    unsigned int i;
+
+    // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+    addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
+    filtersReg = _mm_loadu_si128((__m128i *)filter);
+    // converting the 16 bit (short) to  8 bit (byte) and have the same data
+    // in both lanes of 128 bit register.
+    filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+
+    // duplicate only the first 16 bits in the filter
+    firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u));
+    // duplicate only the second 16 bits in the filter
+    secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
+    // duplicate only the third 16 bits in the filter
+    thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
+    // duplicate only the forth 16 bits in the filter
+    forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
+
+
+    for (i = 0; i < output_height; i++) {
+        // load the first 16 bytes
+        srcRegFilt1 = _mm_loadu_si128((__m128i *)(src_ptr));
+        // load the next 16 bytes in stride of src_pitch
+        srcRegFilt2 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch));
+        srcRegFilt3 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch*6));
+        srcRegFilt4 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch*7));
+
+        // merge the result together
+        srcRegFilt5 = _mm_unpacklo_epi8(srcRegFilt1, srcRegFilt2);
+        srcRegFilt6 = _mm_unpacklo_epi8(srcRegFilt3, srcRegFilt4);
+        srcRegFilt1 = _mm_unpackhi_epi8(srcRegFilt1, srcRegFilt2);
+        srcRegFilt3 = _mm_unpackhi_epi8(srcRegFilt3, srcRegFilt4);
+
+        // multiply 2 adjacent elements with the filter and add the result
+        srcRegFilt5 = _mm_maddubs_epi16(srcRegFilt5, firstFilters);
+        srcRegFilt6 = _mm_maddubs_epi16(srcRegFilt6, forthFilters);
+        srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
+        srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, forthFilters);
+
+
+        // add and saturate the results together
+        srcRegFilt5 = _mm_adds_epi16(srcRegFilt5, srcRegFilt6);
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt3);
+
+        // load the next 16 bytes in stride of two/three src_pitch
+        srcRegFilt2 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch*2));
+        srcRegFilt3 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch*3));
+
+        // merge the result together
+        srcRegFilt4 = _mm_unpacklo_epi8(srcRegFilt2, srcRegFilt3);
+        srcRegFilt6 = _mm_unpackhi_epi8(srcRegFilt2, srcRegFilt3);
+
+        // multiply 2 adjacent elements with the filter and add the result
+        srcRegFilt4 = _mm_maddubs_epi16(srcRegFilt4, secondFilters);
+        srcRegFilt6 = _mm_maddubs_epi16(srcRegFilt6, secondFilters);
+
+        // load the next 16 bytes in stride of four/five src_pitch
+        srcRegFilt2 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch*4));
+        srcRegFilt3 = _mm_loadu_si128((__m128i *)(src_ptr+src_pitch*5));
+
+        // merge the result together
+        srcRegFilt7 = _mm_unpacklo_epi8(srcRegFilt2, srcRegFilt3);
+        srcRegFilt8 = _mm_unpackhi_epi8(srcRegFilt2, srcRegFilt3);
+
+        // multiply 2 adjacent elements with the filter and add the result
+        srcRegFilt7 = _mm_maddubs_epi16(srcRegFilt7, thirdFilters);
+        srcRegFilt8 = _mm_maddubs_epi16(srcRegFilt8, thirdFilters);
+
+
+        // add and saturate the results together
+        srcRegFilt5 = _mm_adds_epi16(srcRegFilt5,
+        _mm_min_epi16(srcRegFilt4, srcRegFilt7));
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
+        _mm_min_epi16(srcRegFilt6, srcRegFilt8));
+
+
+        // add and saturate the results together
+        srcRegFilt5 = _mm_adds_epi16(srcRegFilt5,
+        _mm_max_epi16(srcRegFilt4, srcRegFilt7));
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
+        _mm_max_epi16(srcRegFilt6, srcRegFilt8));
+        srcRegFilt5 = _mm_adds_epi16(srcRegFilt5, addFilterReg64);
+        srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
+
+        // shift by 7 bit each 16 bit
+        srcRegFilt5 = _mm_srai_epi16(srcRegFilt5, 7);
+        srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
+
+        // shrink to 8 bit each 16 bits, the first lane contain the first
+        // convolve result and the second lane contain the second convolve
+        // result
+        srcRegFilt1 = _mm_packus_epi16(srcRegFilt5, srcRegFilt1);
+
+        src_ptr+=src_pitch;
+
+        // save 16 bytes convolve result
+        _mm_store_si128((__m128i*)output_ptr, srcRegFilt1);
+
+        output_ptr+=out_pitch;
+    }
+}
diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk
index eefbd1ac9..8a8b63d97 100644
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -77,6 +77,7 @@ VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_postproc.c
 VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_loopfilter_mmx.asm
 VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_subpixel_8t_sse2.asm
 VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_subpixel_8t_ssse3.asm
+VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_subpixel_8t_intrin_ssse3.c
 ifeq ($(CONFIG_VP9_POSTPROC),yes)
 VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_postproc_mmx.asm
 VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_postproc_sse2.asm