Merge changes Id6a8c549,Ib1e0650b,Ic369dd86

* changes:
  Refactor x86/vpx_subpixel_8t_intrin_ssse3.c
  Add vpx_dsp/x86/mem_sse2.h
  Add transpose_8bit_{4x4,8x8}() x86 optimization

5 files changed, 419 insertions, 491 deletions

diff --git a/vpx_dsp/x86/convolve_ssse3.h b/vpx_dsp/x86/convolve_ssse3.h
new file mode 100644
index 000000000..b71da0e4e
--- /dev/null
+++ b/vpx_dsp/x86/convolve_ssse3.h
@@ -0,0 +1,48 @@
+/*
+ *  Copyright (c) 2017 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.
+ */
+
+#ifndef VPX_DSP_X86_CONVOLVE_SSSE3_H_
+#define VPX_DSP_X86_CONVOLVE_SSSE3_H_
+
+#include <tmmintrin.h>  // SSSE3
+
+#include "./vpx_config.h"
+
+static INLINE void shuffle_filter_ssse3(const int16_t *const filter,
+                                        __m128i *const f) {
+  const __m128i f_values = _mm_load_si128((const __m128i *)filter);
+  // pack and duplicate the filter values
+  f[0] = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u));
+  f[1] = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u));
+  f[2] = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u));
+  f[3] = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu));
+}
+
+static INLINE __m128i convolve8_8_ssse3(const __m128i *const s,
+                                        const __m128i *const f) {
+  // multiply 2 adjacent elements with the filter and add the result
+  const __m128i k_64 = _mm_set1_epi16(1 << 6);
+  const __m128i x0 = _mm_maddubs_epi16(s[0], f[0]);
+  const __m128i x1 = _mm_maddubs_epi16(s[1], f[1]);
+  const __m128i x2 = _mm_maddubs_epi16(s[2], f[2]);
+  const __m128i x3 = _mm_maddubs_epi16(s[3], f[3]);
+  // add and saturate the results together
+  const __m128i min_x2x1 = _mm_min_epi16(x2, x1);
+  const __m128i max_x2x1 = _mm_max_epi16(x2, x1);
+  __m128i temp = _mm_adds_epi16(x0, x3);
+  temp = _mm_adds_epi16(temp, min_x2x1);
+  temp = _mm_adds_epi16(temp, max_x2x1);
+  // round and shift by 7 bit each 16 bit
+  temp = _mm_adds_epi16(temp, k_64);
+  temp = _mm_srai_epi16(temp, 7);
+  return temp;
+}
+
+#endif  // VPX_DSP_X86_CONVOLVE_SSSE3_H_
diff --git a/vpx_dsp/x86/mem_sse2.h b/vpx_dsp/x86/mem_sse2.h
new file mode 100644
index 000000000..24e298d71
--- /dev/null
+++ b/vpx_dsp/x86/mem_sse2.h
@@ -0,0 +1,90 @@
+/*
+ *  Copyright (c) 2017 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.
+ */
+
+#ifndef VPX_DSP_X86_MEM_SSE2_H_
+#define VPX_DSP_X86_MEM_SSE2_H_
+
+#include <emmintrin.h>  // SSE2
+
+#include "./vpx_config.h"
+
+static INLINE void load_8bit_4x4(const uint8_t *const s, const ptrdiff_t stride,
+                                 __m128i *const d) {
+  d[0] = _mm_cvtsi32_si128(*(const int *)(s + 0 * stride));
+  d[1] = _mm_cvtsi32_si128(*(const int *)(s + 1 * stride));
+  d[2] = _mm_cvtsi32_si128(*(const int *)(s + 2 * stride));
+  d[3] = _mm_cvtsi32_si128(*(const int *)(s + 3 * stride));
+}
+
+static INLINE void load_8bit_4x8(const uint8_t *const s, const ptrdiff_t stride,
+                                 __m128i *const d) {
+  load_8bit_4x4(s + 0 * stride, stride, &d[0]);
+  load_8bit_4x4(s + 4 * stride, stride, &d[4]);
+}
+
+static INLINE void load_8bit_8x4(const uint8_t *const s, const ptrdiff_t stride,
+                                 __m128i *const d) {
+  d[0] = _mm_loadl_epi64((const __m128i *)(s + 0 * stride));
+  d[1] = _mm_loadl_epi64((const __m128i *)(s + 1 * stride));
+  d[2] = _mm_loadl_epi64((const __m128i *)(s + 2 * stride));
+  d[3] = _mm_loadl_epi64((const __m128i *)(s + 3 * stride));
+}
+
+static INLINE void load_8bit_8x8(const uint8_t *const s, const ptrdiff_t stride,
+                                 __m128i *const d) {
+  load_8bit_8x4(s + 0 * stride, stride, &d[0]);
+  load_8bit_8x4(s + 4 * stride, stride, &d[4]);
+}
+
+static INLINE void load_8bit_16x8(const uint8_t *const s,
+                                  const ptrdiff_t stride, __m128i *const d) {
+  d[0] = _mm_load_si128((const __m128i *)(s + 0 * stride));
+  d[1] = _mm_load_si128((const __m128i *)(s + 1 * stride));
+  d[2] = _mm_load_si128((const __m128i *)(s + 2 * stride));
+  d[3] = _mm_load_si128((const __m128i *)(s + 3 * stride));
+  d[4] = _mm_load_si128((const __m128i *)(s + 4 * stride));
+  d[5] = _mm_load_si128((const __m128i *)(s + 5 * stride));
+  d[6] = _mm_load_si128((const __m128i *)(s + 6 * stride));
+  d[7] = _mm_load_si128((const __m128i *)(s + 7 * stride));
+}
+
+static INLINE void loadu_8bit_16x8(const uint8_t *const s,
+                                   const ptrdiff_t stride, __m128i *const d) {
+  d[0] = _mm_loadu_si128((const __m128i *)(s + 0 * stride));
+  d[1] = _mm_loadu_si128((const __m128i *)(s + 1 * stride));
+  d[2] = _mm_loadu_si128((const __m128i *)(s + 2 * stride));
+  d[3] = _mm_loadu_si128((const __m128i *)(s + 3 * stride));
+  d[4] = _mm_loadu_si128((const __m128i *)(s + 4 * stride));
+  d[5] = _mm_loadu_si128((const __m128i *)(s + 5 * stride));
+  d[6] = _mm_loadu_si128((const __m128i *)(s + 6 * stride));
+  d[7] = _mm_loadu_si128((const __m128i *)(s + 7 * stride));
+}
+
+static INLINE void store_8bit_4x4(const __m128i *const s, uint8_t *const d,
+                                  const ptrdiff_t stride) {
+  *(int *)(d + 0 * stride) = _mm_cvtsi128_si32(s[0]);
+  *(int *)(d + 1 * stride) = _mm_cvtsi128_si32(s[1]);
+  *(int *)(d + 2 * stride) = _mm_cvtsi128_si32(s[2]);
+  *(int *)(d + 3 * stride) = _mm_cvtsi128_si32(s[3]);
+}
+
+static INLINE void store_8bit_8x8(const __m128i *const s, uint8_t *const d,
+                                  const ptrdiff_t stride) {
+  _mm_storel_epi64((__m128i *)(d + 0 * stride), s[0]);
+  _mm_storel_epi64((__m128i *)(d + 1 * stride), s[1]);
+  _mm_storel_epi64((__m128i *)(d + 2 * stride), s[2]);
+  _mm_storel_epi64((__m128i *)(d + 3 * stride), s[3]);
+  _mm_storel_epi64((__m128i *)(d + 4 * stride), s[4]);
+  _mm_storel_epi64((__m128i *)(d + 5 * stride), s[5]);
+  _mm_storel_epi64((__m128i *)(d + 6 * stride), s[6]);
+  _mm_storel_epi64((__m128i *)(d + 7 * stride), s[7]);
+}
+
+#endif  // VPX_DSP_X86_MEM_SSE2_H_
diff --git a/vpx_dsp/x86/transpose_sse2.h b/vpx_dsp/x86/transpose_sse2.h
index 067103a58..8a0119ca7 100644
--- a/vpx_dsp/x86/transpose_sse2.h
+++ b/vpx_dsp/x86/transpose_sse2.h
@@ -11,7 +11,86 @@
 #ifndef VPX_DSP_X86_TRANSPOSE_SSE2_H_
 #define VPX_DSP_X86_TRANSPOSE_SSE2_H_
 
-#include "./vpx_dsp_rtcd.h"
+#include <emmintrin.h>  // SSE2
+
+#include "./vpx_config.h"
+
+static INLINE __m128i transpose_8bit_4x4(const __m128i *const in) {
+  // Unpack 16 bit elements. Goes from:
+  // in[0]: 00 01 02 03
+  // in[1]: 10 11 12 13
+  // in[2]: 20 21 22 23
+  // in[3]: 30 31 32 33
+  // to:
+  // a0:    00 10 01 11  02 12 03 13
+  // a1:    20 30 21 31  22 32 23 33
+  const __m128i a0 = _mm_unpacklo_epi8(in[0], in[1]);
+  const __m128i a1 = _mm_unpacklo_epi8(in[2], in[3]);
+
+  // Unpack 32 bit elements resulting in:
+  // 00 10 20 30  01 11 21 31  02 12 22 32  03 13 23 33
+  return _mm_unpacklo_epi16(a0, a1);
+}
+
+static INLINE void transpose_8bit_8x8(const __m128i *const in,
+                                      __m128i *const out) {
+  // Unpack 8 bit elements. Goes from:
+  // in[0]: 00 01 02 03 04 05 06 07
+  // in[1]: 10 11 12 13 14 15 16 17
+  // in[2]: 20 21 22 23 24 25 26 27
+  // in[3]: 30 31 32 33 34 35 36 37
+  // in[4]: 40 41 42 43 44 45 46 47
+  // in[5]: 50 51 52 53 54 55 56 57
+  // in[6]: 60 61 62 63 64 65 66 67
+  // in[7]: 70 71 72 73 74 75 76 77
+  // to:
+  // a0:    00 10 01 11 02 12 03 13  04 14 05 15 06 16 07 17
+  // a1:    20 30 21 31 22 32 23 33  24 34 25 35 26 36 27 37
+  // a2:    40 50 41 51 42 52 43 53  44 54 45 55 46 56 47 57
+  // a3:    60 70 61 71 62 72 63 73  64 74 65 75 66 76 67 77
+  const __m128i a0 = _mm_unpacklo_epi8(in[0], in[1]);
+  const __m128i a1 = _mm_unpacklo_epi8(in[2], in[3]);
+  const __m128i a2 = _mm_unpacklo_epi8(in[4], in[5]);
+  const __m128i a3 = _mm_unpacklo_epi8(in[6], in[7]);
+
+  // Unpack 16 bit elements resulting in:
+  // b0: 00 10 20 30 01 11 21 31  02 12 22 32 03 13 23 33
+  // b1: 40 50 60 70 41 51 61 71  42 52 62 72 43 53 63 73
+  // b2: 04 14 24 34 05 15 25 35  06 16 26 36 07 17 27 37
+  // b3: 44 54 64 74 45 55 65 75  46 56 66 76 47 57 67 77
+  const __m128i b0 = _mm_unpacklo_epi16(a0, a1);
+  const __m128i b1 = _mm_unpackhi_epi16(a0, a1);
+  const __m128i b2 = _mm_unpacklo_epi16(a2, a3);
+  const __m128i b3 = _mm_unpackhi_epi16(a2, a3);
+
+  // Unpack 32 bit elements resulting in:
+  // c0: 00 10 20 30 40 50 60 70  01 11 21 31 41 51 61 71
+  // c1: 02 12 22 32 42 52 62 72  03 13 23 33 43 53 63 73
+  // c2: 04 14 24 34 44 54 64 74  05 15 25 35 45 55 65 75
+  // c3: 06 16 26 36 46 56 66 76  07 17 27 37 47 57 67 77
+  const __m128i c0 = _mm_unpacklo_epi32(b0, b2);
+  const __m128i c1 = _mm_unpackhi_epi32(b0, b2);
+  const __m128i c2 = _mm_unpacklo_epi32(b1, b3);
+  const __m128i c3 = _mm_unpackhi_epi32(b1, b3);
+
+  // Unpack 64 bit elements resulting in:
+  // out[0]: 00 10 20 30 40 50 60 70
+  // out[1]: 01 11 21 31 41 51 61 71
+  // out[2]: 02 12 22 32 42 52 62 72
+  // out[3]: 03 13 23 33 43 53 63 73
+  // out[4]: 04 14 24 34 44 54 64 74
+  // out[5]: 05 15 25 35 45 55 65 75
+  // out[6]: 06 16 26 36 46 56 66 76
+  // out[7]: 07 17 27 37 47 57 67 77
+  out[0] = _mm_unpacklo_epi64(c0, c0);
+  out[1] = _mm_unpackhi_epi64(c0, c0);
+  out[2] = _mm_unpacklo_epi64(c1, c1);
+  out[3] = _mm_unpackhi_epi64(c1, c1);
+  out[4] = _mm_unpacklo_epi64(c2, c2);
+  out[5] = _mm_unpackhi_epi64(c2, c2);
+  out[6] = _mm_unpacklo_epi64(c3, c3);
+  out[7] = _mm_unpackhi_epi64(c3, c3);
+}
 
 static INLINE void transpose_16bit_4x4(const __m128i *const in,
                                        __m128i *const out) {
diff --git a/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c b/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c
index 9ffe7d01b..5a94c69b5 100644
--- a/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c
+++ b/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c
@@ -8,50 +8,34 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-#include <tmmintrin.h>
+#include <tmmintrin.h>  // SSSE3
 
 #include <string.h>
 
 #include "./vpx_dsp_rtcd.h"
 #include "vpx_dsp/vpx_filter.h"
 #include "vpx_dsp/x86/convolve.h"
+#include "vpx_dsp/x86/convolve_ssse3.h"
+#include "vpx_dsp/x86/mem_sse2.h"
+#include "vpx_dsp/x86/transpose_sse2.h"
 #include "vpx_mem/vpx_mem.h"
 #include "vpx_ports/mem.h"
 #include "vpx_ports/emmintrin_compat.h"
 
-// filters only for the 4_h8 convolution
-DECLARE_ALIGNED(16, static const uint8_t, filt1_4_h8[16]) = {
-  0, 1, 1, 2, 2, 3, 3, 4, 2, 3, 3, 4, 4, 5, 5, 6
-};
-
-DECLARE_ALIGNED(16, static const uint8_t, 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, static const uint8_t, filt1_global[16]) = {
-  0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8
-};
-
-DECLARE_ALIGNED(16, static const uint8_t, filt2_global[16]) = {
-  2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10
-};
-
-DECLARE_ALIGNED(16, static const uint8_t, filt3_global[16]) = {
-  4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12
-};
-
-DECLARE_ALIGNED(16, static const uint8_t, filt4_global[16]) = {
-  6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14
-};
-
 // These are reused by the avx2 intrinsics.
-filter8_1dfunction vpx_filter_block1d8_v8_intrin_ssse3;
-filter8_1dfunction vpx_filter_block1d8_h8_intrin_ssse3;
-filter8_1dfunction vpx_filter_block1d4_h8_intrin_ssse3;
+// vpx_filter_block1d8_v8_intrin_ssse3()
+// vpx_filter_block1d8_h8_intrin_ssse3()
+// vpx_filter_block1d4_h8_intrin_ssse3()
+
+static INLINE __m128i shuffle_filter_convolve8_8_ssse3(
+    const __m128i *const s, const int16_t *const filter) {
+  __m128i f[4];
+  shuffle_filter_ssse3(filter, f);
+  return convolve8_8_ssse3(s, f);
+}
 
 void vpx_filter_block1d4_h8_intrin_ssse3(
-    const uint8_t *src_ptr, ptrdiff_t src_pixels_per_line, uint8_t *output_ptr,
+    const uint8_t *src_ptr, ptrdiff_t src_pitch, uint8_t *output_ptr,
     ptrdiff_t output_pitch, uint32_t output_height, const int16_t *filter) {
   __m128i firstFilters, secondFilters, shuffle1, shuffle2;
   __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt4;
@@ -77,8 +61,8 @@ void vpx_filter_block1d4_h8_intrin_ssse3(
   secondFilters = _mm_shufflehi_epi16(secondFilters, 0xFFu);
 
   // loading the local filters
-  shuffle1 = _mm_load_si128((__m128i const *)filt1_4_h8);
-  shuffle2 = _mm_load_si128((__m128i const *)filt2_4_h8);
+  shuffle1 = _mm_setr_epi8(0, 1, 1, 2, 2, 3, 3, 4, 2, 3, 3, 4, 4, 5, 5, 6);
+  shuffle2 = _mm_setr_epi8(4, 5, 5, 6, 6, 7, 7, 8, 6, 7, 7, 8, 8, 9, 9, 10);
 
   for (i = 0; i < output_height; i++) {
     srcReg = _mm_loadu_si128((const __m128i *)(src_ptr - 3));
@@ -109,7 +93,7 @@ void vpx_filter_block1d4_h8_intrin_ssse3(
 
     // shrink to 8 bit each 16 bits
     srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
-    src_ptr += src_pixels_per_line;
+    src_ptr += src_pitch;
 
     // save only 4 bytes
     *((int *)&output_ptr[0]) = _mm_cvtsi128_si32(srcRegFilt1);
@@ -119,77 +103,35 @@ void vpx_filter_block1d4_h8_intrin_ssse3(
 }
 
 void vpx_filter_block1d8_h8_intrin_ssse3(
-    const uint8_t *src_ptr, ptrdiff_t src_pixels_per_line, uint8_t *output_ptr,
+    const uint8_t *src_ptr, ptrdiff_t src_pitch, uint8_t *output_ptr,
     ptrdiff_t output_pitch, uint32_t output_height, const 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;
+  __m128i f[4], filt[4], s[4];
 
-  // 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((const __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);
+  shuffle_filter_ssse3(filter, f);
+  filt[0] = _mm_setr_epi8(0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8);
+  filt[1] = _mm_setr_epi8(2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10);
+  filt[2] = _mm_setr_epi8(4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12);
+  filt[3] =
+      _mm_setr_epi8(6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14);
 
   for (i = 0; i < output_height; i++) {
-    srcReg = _mm_loadu_si128((const __m128i *)(src_ptr - 3));
+    const __m128i srcReg = _mm_loadu_si128((const __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(srcRegFilt2, srcRegFilt3);
-    srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt4);
-
-    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 bits
-    srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
+    s[0] = _mm_shuffle_epi8(srcReg, filt[0]);
+    s[1] = _mm_shuffle_epi8(srcReg, filt[1]);
+    s[2] = _mm_shuffle_epi8(srcReg, filt[2]);
+    s[3] = _mm_shuffle_epi8(srcReg, filt[3]);
+    s[0] = convolve8_8_ssse3(s, f);
 
     // shrink to 8 bit each 16 bits
-    srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
+    s[0] = _mm_packus_epi16(s[0], s[0]);
 
-    src_ptr += src_pixels_per_line;
+    src_ptr += src_pitch;
 
     // save only 8 bytes
-    _mm_storel_epi64((__m128i *)&output_ptr[0], srcRegFilt1);
+    _mm_storel_epi64((__m128i *)&output_ptr[0], s[0]);
 
     output_ptr += output_pitch;
   }
@@ -198,83 +140,49 @@ void vpx_filter_block1d8_h8_intrin_ssse3(
 void vpx_filter_block1d8_v8_intrin_ssse3(
     const uint8_t *src_ptr, ptrdiff_t src_pitch, uint8_t *output_ptr,
     ptrdiff_t out_pitch, uint32_t output_height, const int16_t *filter) {
-  __m128i addFilterReg64, filtersReg, minReg;
-  __m128i firstFilters, secondFilters, thirdFilters, forthFilters;
-  __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt5;
-  __m128i srcReg1, srcReg2, srcReg3, srcReg4, srcReg5, srcReg6, srcReg7;
-  __m128i srcReg8;
   unsigned int i;
+  __m128i f[4], s[8], ss[4];
 
-  // 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((const __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));
+  shuffle_filter_ssse3(filter, f);
 
   // load the first 7 rows of 8 bytes
-  srcReg1 = _mm_loadl_epi64((const __m128i *)src_ptr);
-  srcReg2 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch));
-  srcReg3 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 2));
-  srcReg4 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 3));
-  srcReg5 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 4));
-  srcReg6 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 5));
-  srcReg7 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 6));
+  s[0] = _mm_loadl_epi64((const __m128i *)(src_ptr + 0 * src_pitch));
+  s[1] = _mm_loadl_epi64((const __m128i *)(src_ptr + 1 * src_pitch));
+  s[2] = _mm_loadl_epi64((const __m128i *)(src_ptr + 2 * src_pitch));
+  s[3] = _mm_loadl_epi64((const __m128i *)(src_ptr + 3 * src_pitch));
+  s[4] = _mm_loadl_epi64((const __m128i *)(src_ptr + 4 * src_pitch));
+  s[5] = _mm_loadl_epi64((const __m128i *)(src_ptr + 5 * src_pitch));
+  s[6] = _mm_loadl_epi64((const __m128i *)(src_ptr + 6 * src_pitch));
 
   for (i = 0; i < output_height; i++) {
     // load the last 8 bytes
-    srcReg8 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 7));
+    s[7] = _mm_loadl_epi64((const __m128i *)(src_ptr + 7 * src_pitch));
 
     // merge the result together
-    srcRegFilt1 = _mm_unpacklo_epi8(srcReg1, srcReg2);
-    srcRegFilt3 = _mm_unpacklo_epi8(srcReg3, srcReg4);
+    ss[0] = _mm_unpacklo_epi8(s[0], s[1]);
+    ss[1] = _mm_unpacklo_epi8(s[2], s[3]);
 
     // merge the result together
-    srcRegFilt2 = _mm_unpacklo_epi8(srcReg5, srcReg6);
-    srcRegFilt5 = _mm_unpacklo_epi8(srcReg7, srcReg8);
-
-    // 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);
+    ss[2] = _mm_unpacklo_epi8(s[4], s[5]);
+    ss[3] = _mm_unpacklo_epi8(s[6], s[7]);
 
+    ss[0] = convolve8_8_ssse3(ss, f);
     // shrink to 8 bit each 16 bits
-    srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
+    ss[0] = _mm_packus_epi16(ss[0], ss[0]);
 
     src_ptr += src_pitch;
 
     // shift down a row
-    srcReg1 = srcReg2;
-    srcReg2 = srcReg3;
-    srcReg3 = srcReg4;
-    srcReg4 = srcReg5;
-    srcReg5 = srcReg6;
-    srcReg6 = srcReg7;
-    srcReg7 = srcReg8;
+    s[0] = s[1];
+    s[1] = s[2];
+    s[2] = s[3];
+    s[3] = s[4];
+    s[4] = s[5];
+    s[5] = s[6];
+    s[6] = s[7];
 
     // save only 8 bytes convolve result
-    _mm_storel_epi64((__m128i *)&output_ptr[0], srcRegFilt1);
+    _mm_storel_epi64((__m128i *)&output_ptr[0], ss[0]);
 
     output_ptr += out_pitch;
   }
@@ -331,120 +239,41 @@ FUN_CONV_1D(vert, y0_q4, y_step_q4, v, src - src_stride * 3, , ssse3);
 FUN_CONV_1D(avg_horiz, x0_q4, x_step_q4, h, src, avg_, ssse3);
 FUN_CONV_1D(avg_vert, y0_q4, y_step_q4, v, src - src_stride * 3, avg_, ssse3);
 
-#define TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, \
-                      out2, out3, out4, out5, out6, out7)                 \
-  {                                                                       \
-    const __m128i tr0_0 = _mm_unpacklo_epi8(in0, in1);                    \
-    const __m128i tr0_1 = _mm_unpacklo_epi8(in2, in3);                    \
-    const __m128i tr0_2 = _mm_unpacklo_epi8(in4, in5);                    \
-    const __m128i tr0_3 = _mm_unpacklo_epi8(in6, in7);                    \
-                                                                          \
-    const __m128i tr1_0 = _mm_unpacklo_epi16(tr0_0, tr0_1);               \
-    const __m128i tr1_1 = _mm_unpackhi_epi16(tr0_0, tr0_1);               \
-    const __m128i tr1_2 = _mm_unpacklo_epi16(tr0_2, tr0_3);               \
-    const __m128i tr1_3 = _mm_unpackhi_epi16(tr0_2, tr0_3);               \
-                                                                          \
-    const __m128i tr2_0 = _mm_unpacklo_epi32(tr1_0, tr1_2);               \
-    const __m128i tr2_1 = _mm_unpackhi_epi32(tr1_0, tr1_2);               \
-    const __m128i tr2_2 = _mm_unpacklo_epi32(tr1_1, tr1_3);               \
-    const __m128i tr2_3 = _mm_unpackhi_epi32(tr1_1, tr1_3);               \
-                                                                          \
-    out0 = _mm_unpacklo_epi64(tr2_0, tr2_0);                              \
-    out1 = _mm_unpackhi_epi64(tr2_0, tr2_0);                              \
-    out2 = _mm_unpacklo_epi64(tr2_1, tr2_1);                              \
-    out3 = _mm_unpackhi_epi64(tr2_1, tr2_1);                              \
-    out4 = _mm_unpacklo_epi64(tr2_2, tr2_2);                              \
-    out5 = _mm_unpackhi_epi64(tr2_2, tr2_2);                              \
-    out6 = _mm_unpacklo_epi64(tr2_3, tr2_3);                              \
-    out7 = _mm_unpackhi_epi64(tr2_3, tr2_3);                              \
-  }
-
-static void filter_horiz_w8_ssse3(const uint8_t *src_x, ptrdiff_t src_pitch,
-                                  uint8_t *dst, const int16_t *x_filter) {
-  const __m128i k_256 = _mm_set1_epi16(1 << 8);
-  const __m128i f_values = _mm_load_si128((const __m128i *)x_filter);
-  // pack and duplicate the filter values
-  const __m128i f1f0 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u));
-  const __m128i f3f2 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u));
-  const __m128i f5f4 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u));
-  const __m128i f7f6 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu));
-  const __m128i A = _mm_loadl_epi64((const __m128i *)src_x);
-  const __m128i B = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch));
-  const __m128i C = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch * 2));
-  const __m128i D = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch * 3));
-  const __m128i E = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch * 4));
-  const __m128i F = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch * 5));
-  const __m128i G = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch * 6));
-  const __m128i H = _mm_loadl_epi64((const __m128i *)(src_x + src_pitch * 7));
-  // 00 01 10 11 02 03 12 13 04 05 14 15 06 07 16 17
-  const __m128i tr0_0 = _mm_unpacklo_epi16(A, B);
-  // 20 21 30 31 22 23 32 33 24 25 34 35 26 27 36 37
-  const __m128i tr0_1 = _mm_unpacklo_epi16(C, D);
-  // 40 41 50 51 42 43 52 53 44 45 54 55 46 47 56 57
-  const __m128i tr0_2 = _mm_unpacklo_epi16(E, F);
-  // 60 61 70 71 62 63 72 73 64 65 74 75 66 67 76 77
-  const __m128i tr0_3 = _mm_unpacklo_epi16(G, H);
-  // 00 01 10 11 20 21 30 31 02 03 12 13 22 23 32 33
-  const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1);
-  // 04 05 14 15 24 25 34 35 06 07 16 17 26 27 36 37
-  const __m128i tr1_1 = _mm_unpackhi_epi32(tr0_0, tr0_1);
-  // 40 41 50 51 60 61 70 71 42 43 52 53 62 63 72 73
-  const __m128i tr1_2 = _mm_unpacklo_epi32(tr0_2, tr0_3);
-  // 44 45 54 55 64 65 74 75 46 47 56 57 66 67 76 77
-  const __m128i tr1_3 = _mm_unpackhi_epi32(tr0_2, tr0_3);
-  // 00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71
-  const __m128i s1s0 = _mm_unpacklo_epi64(tr1_0, tr1_2);
-  const __m128i s3s2 = _mm_unpackhi_epi64(tr1_0, tr1_2);
-  const __m128i s5s4 = _mm_unpacklo_epi64(tr1_1, tr1_3);
-  const __m128i s7s6 = _mm_unpackhi_epi64(tr1_1, tr1_3);
-  // multiply 2 adjacent elements with the filter and add the result
-  const __m128i x0 = _mm_maddubs_epi16(s1s0, f1f0);
-  const __m128i x1 = _mm_maddubs_epi16(s3s2, f3f2);
-  const __m128i x2 = _mm_maddubs_epi16(s5s4, f5f4);
-  const __m128i x3 = _mm_maddubs_epi16(s7s6, f7f6);
-  // add and saturate the results together
-  const __m128i min_x2x1 = _mm_min_epi16(x2, x1);
-  const __m128i max_x2x1 = _mm_max_epi16(x2, x1);
-  __m128i temp = _mm_adds_epi16(x0, x3);
-  temp = _mm_adds_epi16(temp, min_x2x1);
-  temp = _mm_adds_epi16(temp, max_x2x1);
-  // round and shift by 7 bit each 16 bit
-  temp = _mm_mulhrs_epi16(temp, k_256);
+static void filter_horiz_w8_ssse3(const uint8_t *const src,
+                                  const ptrdiff_t src_stride,
+                                  uint8_t *const dst,
+                                  const int16_t *const x_filter) {
+  __m128i s[8], ss[4], temp;
+
+  load_8bit_8x8(src, src_stride, s);
+  // 00 01 10 11 20 21 30 31  40 41 50 51 60 61 70 71
+  // 02 03 12 13 22 23 32 33  42 43 52 53 62 63 72 73
+  // 04 05 14 15 24 25 34 35  44 45 54 55 64 65 74 75
+  // 06 07 16 17 26 27 36 37  46 47 56 57 66 67 76 77
+  transpose_16bit_4x8(s, ss);
+  temp = shuffle_filter_convolve8_8_ssse3(ss, x_filter);
   // shrink to 8 bit each 16 bits
   temp = _mm_packus_epi16(temp, temp);
   // save only 8 bytes convolve result
   _mm_storel_epi64((__m128i *)dst, temp);
 }
 
-static void transpose8x8_to_dst(const uint8_t *src, ptrdiff_t src_stride,
-                                uint8_t *dst, ptrdiff_t dst_stride) {
-  __m128i A, B, C, D, E, F, G, H;
-
-  A = _mm_loadl_epi64((const __m128i *)src);
-  B = _mm_loadl_epi64((const __m128i *)(src + src_stride));
-  C = _mm_loadl_epi64((const __m128i *)(src + src_stride * 2));
-  D = _mm_loadl_epi64((const __m128i *)(src + src_stride * 3));
-  E = _mm_loadl_epi64((const __m128i *)(src + src_stride * 4));
-  F = _mm_loadl_epi64((const __m128i *)(src + src_stride * 5));
-  G = _mm_loadl_epi64((const __m128i *)(src + src_stride * 6));
-  H = _mm_loadl_epi64((const __m128i *)(src + src_stride * 7));
-
-  TRANSPOSE_8X8(A, B, C, D, E, F, G, H, A, B, C, D, E, F, G, H);
-
-  _mm_storel_epi64((__m128i *)dst, A);
-  _mm_storel_epi64((__m128i *)(dst + dst_stride * 1), B);
-  _mm_storel_epi64((__m128i *)(dst + dst_stride * 2), C);
-  _mm_storel_epi64((__m128i *)(dst + dst_stride * 3), D);
-  _mm_storel_epi64((__m128i *)(dst + dst_stride * 4), E);
-  _mm_storel_epi64((__m128i *)(dst + dst_stride * 5), F);
-  _mm_storel_epi64((__m128i *)(dst + dst_stride * 6), G);
-  _mm_storel_epi64((__m128i *)(dst + dst_stride * 7), H);
+static void transpose8x8_to_dst(const uint8_t *const src,
+                                const ptrdiff_t src_stride, uint8_t *const dst,
+                                const ptrdiff_t dst_stride) {
+  __m128i s[8];
+
+  load_8bit_8x8(src, src_stride, s);
+  transpose_8bit_8x8(s, s);
+  store_8bit_8x8(s, dst, dst_stride);
 }
 
-static void scaledconvolve_horiz_w8(const uint8_t *src, ptrdiff_t src_stride,
-                                    uint8_t *dst, ptrdiff_t dst_stride,
-                                    const InterpKernel *x_filters, int x0_q4,
-                                    int x_step_q4, int w, int h) {
+static void scaledconvolve_horiz_w8(const uint8_t *src,
+                                    const ptrdiff_t src_stride, uint8_t *dst,
+                                    const ptrdiff_t dst_stride,
+                                    const InterpKernel *const x_filters,
+                                    const int x0_q4, const int x_step_q4,
+                                    const int w, const int h) {
   DECLARE_ALIGNED(16, uint8_t, temp[8 * 8]);
   int x, y, z;
   src -= SUBPEL_TAPS / 2 - 1;
@@ -480,93 +309,50 @@ static void scaledconvolve_horiz_w8(const uint8_t *src, ptrdiff_t src_stride,
   } while (y -= 8);
 }
 
-static void filter_horiz_w4_ssse3(const uint8_t *src_ptr, ptrdiff_t src_pitch,
-                                  uint8_t *dst, const int16_t *filter) {
-  const __m128i k_256 = _mm_set1_epi16(1 << 8);
-  const __m128i f_values = _mm_load_si128((const __m128i *)filter);
-  // pack and duplicate the filter values
-  const __m128i f1f0 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u));
-  const __m128i f3f2 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u));
-  const __m128i f5f4 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u));
-  const __m128i f7f6 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu));
-  const __m128i A = _mm_loadl_epi64((const __m128i *)src_ptr);
-  const __m128i B = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch));
-  const __m128i C = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 2));
-  const __m128i D = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 3));
-  // TRANSPOSE...
-  // 00 01 02 03 04 05 06 07
-  // 10 11 12 13 14 15 16 17
-  // 20 21 22 23 24 25 26 27
-  // 30 31 32 33 34 35 36 37
-  //
-  // TO
-  //
-  // 00 10 20 30
-  // 01 11 21 31
-  // 02 12 22 32
-  // 03 13 23 33
-  // 04 14 24 34
-  // 05 15 25 35
-  // 06 16 26 36
-  // 07 17 27 37
-  //
-  // 00 01 10 11 02 03 12 13 04 05 14 15 06 07 16 17
-  const __m128i tr0_0 = _mm_unpacklo_epi16(A, B);
-  // 20 21 30 31 22 23 32 33 24 25 34 35 26 27 36 37
-  const __m128i tr0_1 = _mm_unpacklo_epi16(C, D);
-  // 00 01 10 11 20 21 30 31 02 03 12 13 22 23 32 33
-  const __m128i s1s0 = _mm_unpacklo_epi32(tr0_0, tr0_1);
-  // 04 05 14 15 24 25 34 35 06 07 16 17 26 27 36 37
-  const __m128i s5s4 = _mm_unpackhi_epi32(tr0_0, tr0_1);
+static void filter_horiz_w4_ssse3(const uint8_t *const src,
+                                  const ptrdiff_t src_stride,
+                                  uint8_t *const dst,
+                                  const int16_t *const filter) {
+  __m128i s[4], ss[2];
+  __m128i temp;
+
+  load_8bit_8x4(src, src_stride, s);
+  transpose_16bit_4x4(s, ss);
+  // 00 01 10 11 20 21 30 31
+  s[0] = ss[0];
   // 02 03 12 13 22 23 32 33
-  const __m128i s3s2 = _mm_srli_si128(s1s0, 8);
+  s[1] = _mm_srli_si128(ss[0], 8);
+  // 04 05 14 15 24 25 34 35
+  s[2] = ss[1];
   // 06 07 16 17 26 27 36 37
-  const __m128i s7s6 = _mm_srli_si128(s5s4, 8);
-  // multiply 2 adjacent elements with the filter and add the result
-  const __m128i x0 = _mm_maddubs_epi16(s1s0, f1f0);
-  const __m128i x1 = _mm_maddubs_epi16(s3s2, f3f2);
-  const __m128i x2 = _mm_maddubs_epi16(s5s4, f5f4);
-  const __m128i x3 = _mm_maddubs_epi16(s7s6, f7f6);
-  // add and saturate the results together
-  const __m128i min_x2x1 = _mm_min_epi16(x2, x1);
-  const __m128i max_x2x1 = _mm_max_epi16(x2, x1);
-  __m128i temp = _mm_adds_epi16(x0, x3);
-  temp = _mm_adds_epi16(temp, min_x2x1);
-  temp = _mm_adds_epi16(temp, max_x2x1);
-  // round and shift by 7 bit each 16 bit
-  temp = _mm_mulhrs_epi16(temp, k_256);
+  s[3] = _mm_srli_si128(ss[1], 8);
+
+  temp = shuffle_filter_convolve8_8_ssse3(s, filter);
   // shrink to 8 bit each 16 bits
   temp = _mm_packus_epi16(temp, temp);
   // save only 4 bytes
   *(int *)dst = _mm_cvtsi128_si32(temp);
 }
 
-static void transpose4x4_to_dst(const uint8_t *src, ptrdiff_t src_stride,
-                                uint8_t *dst, ptrdiff_t dst_stride) {
-  __m128i A = _mm_cvtsi32_si128(*(const int *)src);
-  __m128i B = _mm_cvtsi32_si128(*(const int *)(src + src_stride));
-  __m128i C = _mm_cvtsi32_si128(*(const int *)(src + src_stride * 2));
-  __m128i D = _mm_cvtsi32_si128(*(const int *)(src + src_stride * 3));
-  // 00 10 01 11 02 12 03 13
-  const __m128i tr0_0 = _mm_unpacklo_epi8(A, B);
-  // 20 30 21 31 22 32 23 33
-  const __m128i tr0_1 = _mm_unpacklo_epi8(C, D);
-  // 00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33
-  A = _mm_unpacklo_epi16(tr0_0, tr0_1);
-  B = _mm_srli_si128(A, 4);
-  C = _mm_srli_si128(A, 8);
-  D = _mm_srli_si128(A, 12);
-
-  *(int *)(dst) = _mm_cvtsi128_si32(A);
-  *(int *)(dst + dst_stride) = _mm_cvtsi128_si32(B);
-  *(int *)(dst + dst_stride * 2) = _mm_cvtsi128_si32(C);
-  *(int *)(dst + dst_stride * 3) = _mm_cvtsi128_si32(D);
+static void transpose4x4_to_dst(const uint8_t *const src,
+                                const ptrdiff_t src_stride, uint8_t *const dst,
+                                const ptrdiff_t dst_stride) {
+  __m128i s[4];
+
+  load_8bit_4x4(src, src_stride, s);
+  s[0] = transpose_8bit_4x4(s);
+  s[1] = _mm_srli_si128(s[0], 4);
+  s[2] = _mm_srli_si128(s[0], 8);
+  s[3] = _mm_srli_si128(s[0], 12);
+  store_8bit_4x4(s, dst, dst_stride);
 }
 
-static void scaledconvolve_horiz_w4(const uint8_t *src, ptrdiff_t src_stride,
-                                    uint8_t *dst, ptrdiff_t dst_stride,
-                                    const InterpKernel *x_filters, int x0_q4,
-                                    int x_step_q4, int w, int h) {
+static void scaledconvolve_horiz_w4(const uint8_t *src,
+                                    const ptrdiff_t src_stride, uint8_t *dst,
+                                    const ptrdiff_t dst_stride,
+                                    const InterpKernel *const x_filters,
+                                    const int x0_q4, const int x_step_q4,
+                                    const int w, const int h) {
   DECLARE_ALIGNED(16, uint8_t, temp[4 * 4]);
   int x, y, z;
   src -= SUBPEL_TAPS / 2 - 1;
@@ -598,50 +384,41 @@ static void scaledconvolve_horiz_w4(const uint8_t *src, ptrdiff_t src_stride,
   }
 }
 
-static void filter_vert_w4_ssse3(const uint8_t *src_ptr, ptrdiff_t src_pitch,
-                                 uint8_t *dst, const int16_t *filter) {
-  const __m128i k_256 = _mm_set1_epi16(1 << 8);
-  const __m128i f_values = _mm_load_si128((const __m128i *)filter);
-  // pack and duplicate the filter values
-  const __m128i f1f0 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u));
-  const __m128i f3f2 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u));
-  const __m128i f5f4 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u));
-  const __m128i f7f6 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu));
-  const __m128i A = _mm_cvtsi32_si128(*(const int *)src_ptr);
-  const __m128i B = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch));
-  const __m128i C = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch * 2));
-  const __m128i D = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch * 3));
-  const __m128i E = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch * 4));
-  const __m128i F = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch * 5));
-  const __m128i G = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch * 6));
-  const __m128i H = _mm_cvtsi32_si128(*(const int *)(src_ptr + src_pitch * 7));
-  const __m128i s1s0 = _mm_unpacklo_epi8(A, B);
-  const __m128i s3s2 = _mm_unpacklo_epi8(C, D);
-  const __m128i s5s4 = _mm_unpacklo_epi8(E, F);
-  const __m128i s7s6 = _mm_unpacklo_epi8(G, H);
-  // multiply 2 adjacent elements with the filter and add the result
-  const __m128i x0 = _mm_maddubs_epi16(s1s0, f1f0);
-  const __m128i x1 = _mm_maddubs_epi16(s3s2, f3f2);
-  const __m128i x2 = _mm_maddubs_epi16(s5s4, f5f4);
-  const __m128i x3 = _mm_maddubs_epi16(s7s6, f7f6);
-  // add and saturate the results together
-  const __m128i min_x2x1 = _mm_min_epi16(x2, x1);
-  const __m128i max_x2x1 = _mm_max_epi16(x2, x1);
-  __m128i temp = _mm_adds_epi16(x0, x3);
-  temp = _mm_adds_epi16(temp, min_x2x1);
-  temp = _mm_adds_epi16(temp, max_x2x1);
-  // round and shift by 7 bit each 16 bit
-  temp = _mm_mulhrs_epi16(temp, k_256);
+static __m128i filter_vert_kernel(const __m128i *const s,
+                                  const int16_t *const filter) {
+  __m128i ss[4];
+  __m128i temp;
+
+  // 00 10 01 11 02 12 03 13
+  ss[0] = _mm_unpacklo_epi8(s[0], s[1]);
+  // 20 30 21 31 22 32 23 33
+  ss[1] = _mm_unpacklo_epi8(s[2], s[3]);
+  // 40 50 41 51 42 52 43 53
+  ss[2] = _mm_unpacklo_epi8(s[4], s[5]);
+  // 60 70 61 71 62 72 63 73
+  ss[3] = _mm_unpacklo_epi8(s[6], s[7]);
+
+  temp = shuffle_filter_convolve8_8_ssse3(ss, filter);
   // shrink to 8 bit each 16 bits
-  temp = _mm_packus_epi16(temp, temp);
+  return _mm_packus_epi16(temp, temp);
+}
+
+static void filter_vert_w4_ssse3(const uint8_t *const src,
+                                 const ptrdiff_t src_stride, uint8_t *const dst,
+                                 const int16_t *const filter) {
+  __m128i s[8];
+  __m128i temp;
+
+  load_8bit_4x8(src, src_stride, s);
+  temp = filter_vert_kernel(s, filter);
   // save only 4 bytes
   *(int *)dst = _mm_cvtsi128_si32(temp);
 }
 
-static void scaledconvolve_vert_w4(const uint8_t *src, ptrdiff_t src_stride,
-                                   uint8_t *dst, ptrdiff_t dst_stride,
-                                   const InterpKernel *y_filters, int y0_q4,
-                                   int y_step_q4, int w, int h) {
+static void scaledconvolve_vert_w4(
+    const uint8_t *src, const ptrdiff_t src_stride, uint8_t *const dst,
+    const ptrdiff_t dst_stride, const InterpKernel *const y_filters,
+    const int y0_q4, const int y_step_q4, const int w, const int h) {
   int y;
   int y_q4 = y0_q4;
 
@@ -660,50 +437,21 @@ static void scaledconvolve_vert_w4(const uint8_t *src, ptrdiff_t src_stride,
   }
 }
 
-static void filter_vert_w8_ssse3(const uint8_t *src_ptr, ptrdiff_t src_pitch,
-                                 uint8_t *dst, const int16_t *filter) {
-  const __m128i k_256 = _mm_set1_epi16(1 << 8);
-  const __m128i f_values = _mm_load_si128((const __m128i *)filter);
-  // pack and duplicate the filter values
-  const __m128i f1f0 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u));
-  const __m128i f3f2 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u));
-  const __m128i f5f4 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u));
-  const __m128i f7f6 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu));
-  const __m128i A = _mm_loadl_epi64((const __m128i *)src_ptr);
-  const __m128i B = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch));
-  const __m128i C = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 2));
-  const __m128i D = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 3));
-  const __m128i E = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 4));
-  const __m128i F = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 5));
-  const __m128i G = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 6));
-  const __m128i H = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 7));
-  const __m128i s1s0 = _mm_unpacklo_epi8(A, B);
-  const __m128i s3s2 = _mm_unpacklo_epi8(C, D);
-  const __m128i s5s4 = _mm_unpacklo_epi8(E, F);
-  const __m128i s7s6 = _mm_unpacklo_epi8(G, H);
-  // multiply 2 adjacent elements with the filter and add the result
-  const __m128i x0 = _mm_maddubs_epi16(s1s0, f1f0);
-  const __m128i x1 = _mm_maddubs_epi16(s3s2, f3f2);
-  const __m128i x2 = _mm_maddubs_epi16(s5s4, f5f4);
-  const __m128i x3 = _mm_maddubs_epi16(s7s6, f7f6);
-  // add and saturate the results together
-  const __m128i min_x2x1 = _mm_min_epi16(x2, x1);
-  const __m128i max_x2x1 = _mm_max_epi16(x2, x1);
-  __m128i temp = _mm_adds_epi16(x0, x3);
-  temp = _mm_adds_epi16(temp, min_x2x1);
-  temp = _mm_adds_epi16(temp, max_x2x1);
-  // round and shift by 7 bit each 16 bit
-  temp = _mm_mulhrs_epi16(temp, k_256);
-  // shrink to 8 bit each 16 bits
-  temp = _mm_packus_epi16(temp, temp);
+static void filter_vert_w8_ssse3(const uint8_t *const src,
+                                 const ptrdiff_t src_stride, uint8_t *const dst,
+                                 const int16_t *const filter) {
+  __m128i s[8], temp;
+
+  load_8bit_8x8(src, src_stride, s);
+  temp = filter_vert_kernel(s, filter);
   // save only 8 bytes convolve result
   _mm_storel_epi64((__m128i *)dst, temp);
 }
 
-static void scaledconvolve_vert_w8(const uint8_t *src, ptrdiff_t src_stride,
-                                   uint8_t *dst, ptrdiff_t dst_stride,
-                                   const InterpKernel *y_filters, int y0_q4,
-                                   int y_step_q4, int w, int h) {
+static void scaledconvolve_vert_w8(
+    const uint8_t *src, const ptrdiff_t src_stride, uint8_t *const dst,
+    const ptrdiff_t dst_stride, const InterpKernel *const y_filters,
+    const int y0_q4, const int y_step_q4, const int w, const int h) {
   int y;
   int y_q4 = y0_q4;
 
@@ -720,81 +468,44 @@ static void scaledconvolve_vert_w8(const uint8_t *src, ptrdiff_t src_stride,
   }
 }
 
-static void filter_vert_w16_ssse3(const uint8_t *src_ptr, ptrdiff_t src_pitch,
-                                  uint8_t *dst, const int16_t *filter, int w) {
-  const __m128i k_256 = _mm_set1_epi16(1 << 8);
-  const __m128i f_values = _mm_load_si128((const __m128i *)filter);
-  // pack and duplicate the filter values
-  const __m128i f1f0 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u));
-  const __m128i f3f2 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u));
-  const __m128i f5f4 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u));
-  const __m128i f7f6 = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu));
+static void filter_vert_w16_ssse3(const uint8_t *src,
+                                  const ptrdiff_t src_stride,
+                                  uint8_t *const dst,
+                                  const int16_t *const filter, const int w) {
   int i;
+  __m128i f[4];
+  shuffle_filter_ssse3(filter, f);
 
   for (i = 0; i < w; i += 16) {
-    const __m128i A = _mm_loadu_si128((const __m128i *)src_ptr);
-    const __m128i B = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch));
-    const __m128i C =
-        _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2));
-    const __m128i D =
-        _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3));
-    const __m128i E =
-        _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 4));
-    const __m128i F =
-        _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 5));
-    const __m128i G =
-        _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 6));
-    const __m128i H =
-        _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 7));
-    // merge the result together
-    const __m128i s1s0_lo = _mm_unpacklo_epi8(A, B);
-    const __m128i s7s6_lo = _mm_unpacklo_epi8(G, H);
-    const __m128i s1s0_hi = _mm_unpackhi_epi8(A, B);
-    const __m128i s7s6_hi = _mm_unpackhi_epi8(G, H);
-    // multiply 2 adjacent elements with the filter and add the result
-    const __m128i x0_lo = _mm_maddubs_epi16(s1s0_lo, f1f0);
-    const __m128i x3_lo = _mm_maddubs_epi16(s7s6_lo, f7f6);
-    const __m128i x0_hi = _mm_maddubs_epi16(s1s0_hi, f1f0);
-    const __m128i x3_hi = _mm_maddubs_epi16(s7s6_hi, f7f6);
-    // add and saturate the results together
-    const __m128i x3x0_lo = _mm_adds_epi16(x0_lo, x3_lo);
-    const __m128i x3x0_hi = _mm_adds_epi16(x0_hi, x3_hi);
-    // merge the result together
-    const __m128i s3s2_lo = _mm_unpacklo_epi8(C, D);
-    const __m128i s3s2_hi = _mm_unpackhi_epi8(C, D);
-    // multiply 2 adjacent elements with the filter and add the result
-    const __m128i x1_lo = _mm_maddubs_epi16(s3s2_lo, f3f2);
-    const __m128i x1_hi = _mm_maddubs_epi16(s3s2_hi, f3f2);
+    __m128i s[8], s_lo[4], s_hi[4], temp_lo, temp_hi;
+
+    loadu_8bit_16x8(src, src_stride, s);
+
     // merge the result together
-    const __m128i s5s4_lo = _mm_unpacklo_epi8(E, F);
-    const __m128i s5s4_hi = _mm_unpackhi_epi8(E, F);
-    // multiply 2 adjacent elements with the filter and add the result
-    const __m128i x2_lo = _mm_maddubs_epi16(s5s4_lo, f5f4);
-    const __m128i x2_hi = _mm_maddubs_epi16(s5s4_hi, f5f4);
-    // add and saturate the results together
-    __m128i temp_lo = _mm_adds_epi16(x3x0_lo, _mm_min_epi16(x1_lo, x2_lo));
-    __m128i temp_hi = _mm_adds_epi16(x3x0_hi, _mm_min_epi16(x1_hi, x2_hi));
-
-    // add and saturate the results together
-    temp_lo = _mm_adds_epi16(temp_lo, _mm_max_epi16(x1_lo, x2_lo));
-    temp_hi = _mm_adds_epi16(temp_hi, _mm_max_epi16(x1_hi, x2_hi));
-    // round and shift by 7 bit each 16 bit
-    temp_lo = _mm_mulhrs_epi16(temp_lo, k_256);
-    temp_hi = _mm_mulhrs_epi16(temp_hi, k_256);
-    // shrink to 8 bit each 16 bits, the first lane contain the first
-    // convolve result and the second lane contain the second convolve
-    // result
+    s_lo[0] = _mm_unpacklo_epi8(s[0], s[1]);
+    s_hi[0] = _mm_unpackhi_epi8(s[0], s[1]);
+    s_lo[1] = _mm_unpacklo_epi8(s[2], s[3]);
+    s_hi[1] = _mm_unpackhi_epi8(s[2], s[3]);
+    s_lo[2] = _mm_unpacklo_epi8(s[4], s[5]);
+    s_hi[2] = _mm_unpackhi_epi8(s[4], s[5]);
+    s_lo[3] = _mm_unpacklo_epi8(s[6], s[7]);
+    s_hi[3] = _mm_unpackhi_epi8(s[6], s[7]);
+    temp_lo = convolve8_8_ssse3(s_lo, f);
+    temp_hi = convolve8_8_ssse3(s_hi, f);
+
+    // shrink to 8 bit each 16 bits, the first lane contain the first convolve
+    // result and the second lane contain the second convolve result
     temp_hi = _mm_packus_epi16(temp_lo, temp_hi);
-    src_ptr += 16;
+    src += 16;
     // save 16 bytes convolve result
     _mm_store_si128((__m128i *)&dst[i], temp_hi);
   }
 }
 
-static void scaledconvolve_vert_w16(const uint8_t *src, ptrdiff_t src_stride,
-                                    uint8_t *dst, ptrdiff_t dst_stride,
-                                    const InterpKernel *y_filters, int y0_q4,
-                                    int y_step_q4, int w, int h) {
+static void scaledconvolve_vert_w16(
+    const uint8_t *src, const ptrdiff_t src_stride, uint8_t *const dst,
+    const ptrdiff_t dst_stride, const InterpKernel *const y_filters,
+    const int y0_q4, const int y_step_q4, const int w, const int h) {
   int y;
   int y_q4 = y0_q4;
 
diff --git a/vpx_dsp/x86/vpx_subpixel_8t_ssse3.asm b/vpx_dsp/x86/vpx_subpixel_8t_ssse3.asm
index c1a6f23ab..952d9307d 100644
--- a/vpx_dsp/x86/vpx_subpixel_8t_ssse3.asm
+++ b/vpx_dsp/x86/vpx_subpixel_8t_ssse3.asm
@@ -327,12 +327,12 @@ cglobal filter_block1d16_%1, 6, 6, 14, LOCAL_VARS_SIZE, \
 %endm
 
 INIT_XMM ssse3
-SUBPIX_HFILTER16 h8
-SUBPIX_HFILTER16 h8_avg
-SUBPIX_HFILTER8  h8
-SUBPIX_HFILTER8  h8_avg
-SUBPIX_HFILTER4  h8
-SUBPIX_HFILTER4  h8_avg
+SUBPIX_HFILTER16 h8      ; vpx_filter_block1d16_h8_ssse3
+SUBPIX_HFILTER16 h8_avg  ; vpx_filter_block1d16_h8_avg_ssse3
+SUBPIX_HFILTER8  h8      ; vpx_filter_block1d8_h8_ssse3
+SUBPIX_HFILTER8  h8_avg  ; vpx_filter_block1d8_h8_avg_ssse3
+SUBPIX_HFILTER4  h8      ; vpx_filter_block1d4_h8_ssse3
+SUBPIX_HFILTER4  h8_avg  ; vpx_filter_block1d4_h8_avg_ssse3
 
 ;-------------------------------------------------------------------------------
 
@@ -795,9 +795,9 @@ cglobal filter_block1d16_%1, 6, NUM_GENERAL_REG_USED, 16, LOCAL_VARS_SIZE, \
 %endm
 
 INIT_XMM ssse3
-SUBPIX_VFILTER16     v8
-SUBPIX_VFILTER16 v8_avg
-SUBPIX_VFILTER       v8, 8
-SUBPIX_VFILTER   v8_avg, 8
-SUBPIX_VFILTER       v8, 4
-SUBPIX_VFILTER   v8_avg, 4
+SUBPIX_VFILTER16     v8     ; vpx_filter_block1d16_v8_ssse3
+SUBPIX_VFILTER16 v8_avg     ; vpx_filter_block1d16_v8_avg_ssse3
+SUBPIX_VFILTER       v8, 8  ; vpx_filter_block1d8_v8_ssse3
+SUBPIX_VFILTER   v8_avg, 8  ; vpx_filter_block1d8_v8_avg_ssse3
+SUBPIX_VFILTER       v8, 4  ; vpx_filter_block1d4_v8_ssse3
+SUBPIX_VFILTER   v8_avg, 4  ; vpx_filter_block1d4_v8_avg_ssse3