Use 256-byte aligned filter tables

This avoids duplicating all the filters twice. Includes fixups to the
convolve routines and associated tests to make this work.

Change-Id: I922f86021594e55072ddb63b42b2313605db6e00

4 files changed, 30 insertions, 104 deletions

diff --git a/vp9/common/vp9_convolve.c b/vp9/common/vp9_convolve.c
index f1b5915bd..b062e7dc7 100644
--- a/vp9/common/vp9_convolve.c
+++ b/vp9/common/vp9_convolve.c
@@ -19,7 +19,6 @@
 
 #define VP9_FILTER_WEIGHT 128
 #define VP9_FILTER_SHIFT  7
-#define ALIGN_FILTERS_256 0
 
 /* Assume a bank of 16 filters to choose from. There are two implementations
  * for filter wrapping behavior, since we want to be able to pick which filter
@@ -34,8 +33,11 @@
  *    always 256 byte aligned.
  *
  * Implementations 2 and 3 are likely preferable, as they avoid an extra 2
- * parameters, and switching between them is trivial.
+ * parameters, and switching between them is trivial, with the
+ * ALIGN_FILTERS_256 macro, below.
  */
+ #define ALIGN_FILTERS_256 1
+
 static void convolve_horiz_c(const uint8_t *src, int src_stride,
                              uint8_t *dst, int dst_stride,
                              const int16_t *filter_x0, int x_step_q4,
@@ -56,11 +58,12 @@ static void convolve_horiz_c(const uint8_t *src, int src_stride,
     const int16_t *filter_x = filter_x0;
 
     /* Initial phase offset */
-    int x_q4 = (filter_x - filter_x_base) / taps;
+    int x0_q4 = (filter_x - filter_x_base) / taps;
+    int x_q4 = x0_q4;
 
     for (x = 0; x < w; ++x) {
       /* Per-pixel src offset */
-      int src_x = x_q4 >> 4;
+      int src_x = (x_q4 - x0_q4) >> 4;
 
       for (sum = 0, k = 0; k < taps; ++k) {
         sum += src[src_x + k] * filter_x[k];
@@ -97,11 +100,12 @@ static void convolve_avg_horiz_c(const uint8_t *src, int src_stride,
     const int16_t *filter_x = filter_x0;
 
     /* Initial phase offset */
-    int x_q4 = (filter_x - filter_x_base) / taps;
+    int x0_q4 = (filter_x - filter_x_base) / taps;
+    int x_q4 = x0_q4;
 
     for (x = 0; x < w; ++x) {
       /* Per-pixel src offset */
-      int src_x = x_q4 >> 4;
+      int src_x = (x_q4 - x0_q4) >> 4;
 
       for (sum = 0, k = 0; k < taps; ++k) {
         sum += src[src_x + k] * filter_x[k];
@@ -138,11 +142,12 @@ static void convolve_vert_c(const uint8_t *src, int src_stride,
     const int16_t *filter_y = filter_y0;
 
     /* Initial phase offset */
-    int y_q4 = (filter_y - filter_y_base) / taps;
+    int y0_q4 = (filter_y - filter_y_base) / taps;
+    int y_q4 = y0_q4;
 
     for (y = 0; y < h; ++y) {
       /* Per-pixel src offset */
-      int src_y = y_q4 >> 4;
+      int src_y = (y_q4 - y0_q4) >> 4;
 
       for (sum = 0, k = 0; k < taps; ++k) {
         sum += src[(src_y + k) * src_stride] * filter_y[k];
@@ -179,11 +184,12 @@ static void convolve_avg_vert_c(const uint8_t *src, int src_stride,
     const int16_t *filter_y = filter_y0;
 
     /* Initial phase offset */
-    int y_q4 = (filter_y - filter_y_base) / taps;
+    int y0_q4 = (filter_y - filter_y_base) / taps;
+    int y_q4 = y0_q4;
 
     for (y = 0; y < h; ++y) {
       /* Per-pixel src offset */
-      int src_y = y_q4 >> 4;
+      int src_y = (y_q4 - y0_q4) >> 4;
 
       for (sum = 0, k = 0; k < taps; ++k) {
         sum += src[(src_y + k) * src_stride] * filter_y[k];
diff --git a/vp9/common/vp9_filter.c b/vp9/common/vp9_filter.c
index 9249c5786..434c63e7e 100644
--- a/vp9/common/vp9_filter.c
+++ b/vp9/common/vp9_filter.c
@@ -15,26 +15,7 @@
 #include "vp9_rtcd.h"
 #include "vp9/common/vp9_common.h"
 
-/* TODO(jkoleszar): We can avoid duplicating these tables 2X by forcing 256
- * byte alignment of the table's base address.
- */
-DECLARE_ALIGNED(16, const int16_t, vp9_bilinear_filters[SUBPEL_SHIFTS*2][8]) = {
-  { 0, 0, 0, 128,   0, 0, 0, 0 },
-  { 0, 0, 0, 120,   8, 0, 0, 0 },
-  { 0, 0, 0, 112,  16, 0, 0, 0 },
-  { 0, 0, 0, 104,  24, 0, 0, 0 },
-  { 0, 0, 0,  96,  32, 0, 0, 0 },
-  { 0, 0, 0,  88,  40, 0, 0, 0 },
-  { 0, 0, 0,  80,  48, 0, 0, 0 },
-  { 0, 0, 0,  72,  56, 0, 0, 0 },
-  { 0, 0, 0,  64,  64, 0, 0, 0 },
-  { 0, 0, 0,  56,  72, 0, 0, 0 },
-  { 0, 0, 0,  48,  80, 0, 0, 0 },
-  { 0, 0, 0,  40,  88, 0, 0, 0 },
-  { 0, 0, 0,  32,  96, 0, 0, 0 },
-  { 0, 0, 0,  24, 104, 0, 0, 0 },
-  { 0, 0, 0,  16, 112, 0, 0, 0 },
-  { 0, 0, 0,   8, 120, 0, 0, 0 },
+DECLARE_ALIGNED(256, const int16_t, vp9_bilinear_filters[SUBPEL_SHIFTS][8]) = {
   { 0, 0, 0, 128,   0, 0, 0, 0 },
   { 0, 0, 0, 120,   8, 0, 0, 0 },
   { 0, 0, 0, 112,  16, 0, 0, 0 },
@@ -55,7 +36,7 @@ DECLARE_ALIGNED(16, const int16_t, vp9_bilinear_filters[SUBPEL_SHIFTS*2][8]) = {
 
 #define FILTER_ALPHA       0
 #define FILTER_ALPHA_SHARP 1
-DECLARE_ALIGNED(16, const int16_t, vp9_sub_pel_filters_8[SUBPEL_SHIFTS*2][8])
+DECLARE_ALIGNED(256, const int16_t, vp9_sub_pel_filters_8[SUBPEL_SHIFTS][8])
     = {
 #if FILTER_ALPHA == 0
   /* Lagrangian interpolation filter */
@@ -74,23 +55,8 @@ DECLARE_ALIGNED(16, const int16_t, vp9_sub_pel_filters_8[SUBPEL_SHIFTS*2][8])
   { -1,   4, -11,  37, 112, -16,   4, -1},
   { -1,   3,  -9,  27, 118, -13,   4, -1},
   { 0,   2,  -6,  18, 122, -10,   3, -1},
-  { 0,   1,  -3,   8, 126,  -5,   1,  0},
-  { 0,   0,   0, 128,   0,   0,   0,  0},
-  { 0,   1,  -5, 126,   8,  -3,   1,  0},
-  { -1,   3, -10, 122,  18,  -6,   2,  0},
-  { -1,   4, -13, 118,  27,  -9,   3, -1},
-  { -1,   4, -16, 112,  37, -11,   4, -1},
-  { -1,   5, -18, 105,  48, -14,   4, -1},
-  { -1,   5, -19,  97,  58, -16,   5, -1},
-  { -1,   6, -19,  88,  68, -18,   5, -1},
-  { -1,   6, -19,  78,  78, -19,   6, -1},
-  { -1,   5, -18,  68,  88, -19,   6, -1},
-  { -1,   5, -16,  58,  97, -19,   5, -1},
-  { -1,   4, -14,  48, 105, -18,   5, -1},
-  { -1,   4, -11,  37, 112, -16,   4, -1},
-  { -1,   3,  -9,  27, 118, -13,   4, -1},
-  { 0,   2,  -6,  18, 122, -10,   3, -1},
   { 0,   1,  -3,   8, 126,  -5,   1,  0}
+
 #elif FILTER_ALPHA == 50
   /* Generated using MATLAB:
    * alpha = 0.5;
@@ -118,7 +84,7 @@ DECLARE_ALIGNED(16, const int16_t, vp9_sub_pel_filters_8[SUBPEL_SHIFTS*2][8])
 #endif  /* FILTER_ALPHA */
 };
 
-DECLARE_ALIGNED(16, const int16_t, vp9_sub_pel_filters_8s[SUBPEL_SHIFTS*2][8])
+DECLARE_ALIGNED(256, const int16_t, vp9_sub_pel_filters_8s[SUBPEL_SHIFTS][8])
     = {
 #if FILTER_ALPHA_SHARP == 1
   /* dct based filter */
@@ -137,23 +103,8 @@ DECLARE_ALIGNED(16, const int16_t, vp9_sub_pel_filters_8s[SUBPEL_SHIFTS*2][8])
   {-2,   6, -13,  37, 115, -20,   9, -4},
   {-2,   5, -10,  27, 121, -17,   7, -3},
   {-1,   3,  -6,  17, 125, -13,   5, -2},
-  {0,   1,  -3,   8, 127,  -7,   3, -1},
-  {0,   0,   0, 128,   0,   0,   0, 0},
-  {-1,   3,  -7, 127,   8,  -3,   1, 0},
-  {-2,   5, -13, 125,  17,  -6,   3, -1},
-  {-3,   7, -17, 121,  27, -10,   5, -2},
-  {-4,   9, -20, 115,  37, -13,   6, -2},
-  {-4,  10, -23, 108,  48, -16,   8, -3},
-  {-4,  10, -24, 100,  59, -19,   9, -3},
-  {-4,  11, -24,  90,  70, -21,  10, -4},
-  {-4,  11, -23,  80,  80, -23,  11, -4},
-  {-4,  10, -21,  70,  90, -24,  11, -4},
-  {-3,   9, -19,  59, 100, -24,  10, -4},
-  {-3,   8, -16,  48, 108, -23,  10, -4},
-  {-2,   6, -13,  37, 115, -20,   9, -4},
-  {-2,   5, -10,  27, 121, -17,   7, -3},
-  {-1,   3,  -6,  17, 125, -13,   5, -2},
   {0,   1,  -3,   8, 127,  -7,   3, -1}
+
 #elif FILTER_ALPHA_SHARP == 75
   /* alpha = 0.75 */
   {0,   0,   0, 128,   0,   0,   0, 0},
@@ -175,8 +126,8 @@ DECLARE_ALIGNED(16, const int16_t, vp9_sub_pel_filters_8s[SUBPEL_SHIFTS*2][8])
 #endif  /* FILTER_ALPHA_SHARP */
 };
 
-DECLARE_ALIGNED(16, const int16_t,
-                vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS*2][8]) = {
+DECLARE_ALIGNED(256, const int16_t,
+                vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS][8]) = {
   /* 8-tap lowpass filter */
   /* Hamming window */
   {-1, -7, 32, 80, 32, -7, -1,  0},
@@ -194,26 +145,10 @@ DECLARE_ALIGNED(16, const int16_t,
   { 1, -3, -4, 50, 76, 16, -8,  0},
   { 1, -3, -5, 45, 78, 20, -8,  0},
   { 1, -2, -7, 41, 79, 24, -8,  0},
-  { 1, -2, -7, 37, 80, 28, -8, -1},
-  {-1, -7, 32, 80, 32, -7, -1,  0},
-  {-1, -8, 28, 80, 37, -7, -2,  1},
-  { 0, -8, 24, 79, 41, -7, -2,  1},
-  { 0, -8, 20, 78, 45, -5, -3,  1},
-  { 0, -8, 16, 76, 50, -4, -3,  1},
-  { 0, -7, 13, 74, 54, -3, -4,  1},
-  { 1, -7,  9, 71, 58, -1, -4,  1},
-  { 1, -6,  6, 68, 62,  1, -5,  1},
-  { 1, -6,  4, 65, 65,  4, -6,  1},
-  { 1, -5,  1, 62, 68,  6, -6,  1},
-  { 1, -4, -1, 58, 71,  9, -7,  1},
-  { 1, -4, -3, 54, 74, 13, -7,  0},
-  { 1, -3, -4, 50, 76, 16, -8,  0},
-  { 1, -3, -5, 45, 78, 20, -8,  0},
-  { 1, -2, -7, 41, 79, 24, -8,  0},
   { 1, -2, -7, 37, 80, 28, -8, -1}
 };
 
-DECLARE_ALIGNED(16, const int16_t, vp9_sub_pel_filters_6[SUBPEL_SHIFTS*2][8])
+DECLARE_ALIGNED(256, const int16_t, vp9_sub_pel_filters_6[SUBPEL_SHIFTS][8])
     = {
   {0, 0,   0, 128,   0,   0, 0,  0},
   {0, 1,  -5, 125,   8,  -2, 1,  0},
@@ -230,21 +165,5 @@ DECLARE_ALIGNED(16, const int16_t, vp9_sub_pel_filters_6[SUBPEL_SHIFTS*2][8])
   {0, 2, -10,  37, 110, -14, 3,  0},
   {0, 2,  -8,  27, 116, -11, 2,  0},
   {0, 1,  -5,  17, 122,  -8, 1,  0},
-  {0, 1,  -2,   8, 125,  -5, 1,  0},
-  {0, 0,   0, 128,   0,   0, 0,  0},
-  {0, 1,  -5, 125,   8,  -2, 1,  0},
-  {0, 1,  -8, 122,  17,  -5, 1,  0},
-  {0, 2, -11, 116,  27,  -8, 2,  0},
-  {0, 3, -14, 110,  37, -10, 2,  0},
-  {0, 3, -15, 103,  47, -12, 2,  0},
-  {0, 3, -16,  95,  57, -14, 3,  0},
-  {0, 3, -16,  86,  67, -15, 3,  0},
-  {0, 3, -16,  77,  77, -16, 3,  0},
-  {0, 3, -15,  67,  86, -16, 3,  0},
-  {0, 3, -14,  57,  95, -16, 3,  0},
-  {0, 2, -12,  47, 103, -15, 3,  0},
-  {0, 2, -10,  37, 110, -14, 3,  0},
-  {0, 2,  -8,  27, 116, -11, 2,  0},
-  {0, 1,  -5,  17, 122,  -8, 1,  0},
   {0, 1,  -2,   8, 125,  -5, 1,  0}
 };
diff --git a/vp9/common/vp9_filter.h b/vp9/common/vp9_filter.h
index 765379d35..1ccfdaac2 100644
--- a/vp9/common/vp9_filter.h
+++ b/vp9/common/vp9_filter.h
@@ -21,11 +21,11 @@
 
 #define SUBPEL_SHIFTS 16
 
-extern const int16_t vp9_bilinear_filters[SUBPEL_SHIFTS*2][8];
-extern const int16_t vp9_sub_pel_filters_6[SUBPEL_SHIFTS*2][8];
-extern const int16_t vp9_sub_pel_filters_8[SUBPEL_SHIFTS*2][8];
-extern const int16_t vp9_sub_pel_filters_8s[SUBPEL_SHIFTS*2][8];
-extern const int16_t vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS*2][8];
+extern const int16_t vp9_bilinear_filters[SUBPEL_SHIFTS][8];
+extern const int16_t vp9_sub_pel_filters_6[SUBPEL_SHIFTS][8];
+extern const int16_t vp9_sub_pel_filters_8[SUBPEL_SHIFTS][8];
+extern const int16_t vp9_sub_pel_filters_8s[SUBPEL_SHIFTS][8];
+extern const int16_t vp9_sub_pel_filters_8lp[SUBPEL_SHIFTS][8];
 
 // The VP9_BILINEAR_FILTERS_2TAP macro returns a pointer to the bilinear
 // filter kernel as a 2 tap filter.
diff --git a/vp9/common/vp9_reconinter.c b/vp9/common/vp9_reconinter.c
index 7be0101ac..30e8951af 100644
--- a/vp9/common/vp9_reconinter.c
+++ b/vp9/common/vp9_reconinter.c
@@ -128,6 +128,7 @@ void vp9_setup_interp_filters(MACROBLOCKD *xd,
       break;
 #endif
   }
+  assert(((intptr_t)xd->subpix.filter_x & 0xff) == 0);
 }
 
 void vp9_copy_mem16x16_c(const uint8_t *src,