Change to MV reference search.

This patch reduces the cpu cost of the MV ref
search by only allowing insert for candidates
that would be in the current top 4.

This could alter the outcome and slightly favors
near candidates which are tested first but also
limits the worst case loop count to 4 and means in
many cases it will drop out and not happen.

Change-Id: Idd795a825f9fd681f30f4fcd550c34c38939e113

vp9/common/vp9_blockd.h

@@ -45,6 +45,7 @@ void vpx_log(const char *format, ...);
 #define SEGMENT_DELTADATA   0
 #define SEGMENT_ABSDATA     1
 #define MAX_MV_REFS 9
+#define MAX_MV_REF_CANDIDATES 4
 
 typedef struct {
   int r, c;
@@ -238,7 +239,7 @@ typedef struct {
   MV_REFERENCE_FRAME ref_frame, second_ref_frame;
   TX_SIZE txfm_size;
   int_mv mv[2]; // for each reference frame used
-  int_mv ref_mvs[MAX_REF_FRAMES][MAX_MV_REFS];
+  int_mv ref_mvs[MAX_REF_FRAMES][MAX_MV_REF_CANDIDATES];
   int_mv best_mv, best_second_mv;
 #if CONFIG_NEW_MVREF
   int best_index, best_second_index;

vp9/common/vp9_findnearmv.c

@@ -139,8 +139,8 @@ void vp9_find_best_ref_mvs(MACROBLOCKD *xd,
 #if CONFIG_SUBPELREFMV
   unsigned int sse;
 #endif
-  unsigned int ref_scores[MAX_MV_REFS] = {0};
-  int_mv sorted_mvs[MAX_MV_REFS];
+  unsigned int ref_scores[MAX_MV_REF_CANDIDATES] = {0};
+  int_mv sorted_mvs[MAX_MV_REF_CANDIDATES];
   int zero_seen = FALSE;
 
   // Default all to 0,0 if nothing else available
@@ -159,9 +159,8 @@ void vp9_find_best_ref_mvs(MACROBLOCKD *xd,
   left_ref  = ref_y_buffer - 3;
 #endif
 
-  //for(i = 0; i < MAX_MV_REFS; ++i) {
-  // Limit search to the predicted best 4
-  for(i = 0; i < 4; ++i) {
+  // Limit search to the predicted best few candidates
+  for(i = 0; i < MAX_MV_REF_CANDIDATES; ++i) {
     int_mv this_mv;
     int offset = 0;
     int row_offset, col_offset;
@@ -268,7 +267,7 @@ void vp9_find_best_ref_mvs(MACROBLOCKD *xd,
   }
 
   // Make sure all the candidates are properly clamped etc
-  for (i = 0; i < 4; ++i) {
+  for (i = 0; i < MAX_MV_REF_CANDIDATES; ++i) {
     lower_mv_precision(&sorted_mvs[i], xd->allow_high_precision_mv);
     clamp_mv2(&sorted_mvs[i], xd);
   }

vp9/common/vp9_mvref_common.c

@@ -170,14 +170,20 @@ static void addmv_and_shuffle(
   int weight
 ) {
 
-  int i = *index;
+  int i;
+  int insert_point;
   int duplicate_found = FALSE;
 
-  // Check for duplicates. If there is one increment its score.
-  // Duplicate defined as being the same full pel vector with rounding.
+  // Check for duplicates. If there is one increase its score.
+  // We only compare vs the current top candidates.
+  insert_point = (*index < (MAX_MV_REF_CANDIDATES - 1))
+                 ? *index : (MAX_MV_REF_CANDIDATES - 1);
+
+  i = insert_point;
+  if (*index > i)
+    i++;
   while (i > 0) {
     i--;
-
     if (candidate_mv.as_int == mv_list[i].as_int) {
       duplicate_found = TRUE;
       mv_scores[i] += weight;
@@ -185,11 +191,13 @@ static void addmv_and_shuffle(
     }
   }
 
-  // If no duplicate was found add the new vector and give it a weight
-  if (!duplicate_found) {
-    mv_list[*index].as_int = candidate_mv.as_int;
-    mv_scores[*index] = weight;
-    i = *index;
+  // If no duplicate and the new candidate is good enough then add it.
+  if (!duplicate_found ) {
+    if (weight > mv_scores[insert_point]) {
+      mv_list[insert_point].as_int = candidate_mv.as_int;
+      mv_scores[insert_point] = weight;
+      i = insert_point;
+    }
     (*index)++;
   }
 
@@ -224,12 +232,12 @@ void vp9_find_mv_refs(
   int i;
   MODE_INFO *candidate_mi;
   MB_MODE_INFO * mbmi = &xd->mode_info_context->mbmi;
-  int_mv candidate_mvs[MAX_MV_REFS];
+  int_mv candidate_mvs[MAX_MV_REF_CANDIDATES];
   int_mv c_refmv;
-  MV_REFERENCE_FRAME c_ref_frame;
   int_mv c2_refmv;
+  MV_REFERENCE_FRAME c_ref_frame;
   MV_REFERENCE_FRAME c2_ref_frame;
-  int candidate_scores[MAX_MV_REFS];
+  int candidate_scores[MAX_MV_REF_CANDIDATES];
   int index = 0;
   int split_count = 0;
   int ref_weight = 0;
@@ -238,8 +246,8 @@ void vp9_find_mv_refs(
   int *ref_distance_weight;
 
   // Blank the reference vector lists and other local structures.
-  vpx_memset(mv_ref_list, 0, sizeof(int_mv) * MAX_MV_REFS);
-  vpx_memset(candidate_mvs, 0, sizeof(int_mv) * MAX_MV_REFS);
+  vpx_memset(mv_ref_list, 0, sizeof(int_mv) * MAX_MV_REF_CANDIDATES);
+  vpx_memset(candidate_mvs, 0, sizeof(int_mv) * MAX_MV_REF_CANDIDATES);
   vpx_memset(candidate_scores, 0, sizeof(candidate_scores));
 
 #if CONFIG_SUPERBLOCKS
@@ -349,11 +357,6 @@ void vp9_find_mv_refs(
     }
   }
 
-  // Make sure we are able to add 0,0
-  if (index > (MAX_MV_REFS - 1)) {
-    index = (MAX_MV_REFS - 1);
-  }
-
   // Define inter mode coding context.
   // 0,0 was best
   if (candidate_mvs[0].as_int == 0) {
@@ -383,14 +386,12 @@ void vp9_find_mv_refs(
   }
 
   // 0,0 is always a valid reference.
-  for (i = 0; i < index; ++i) {
+  for (i = 0; i < MAX_MV_REF_CANDIDATES; ++i) {
     if (candidate_mvs[i].as_int == 0)
       break;
   }
-  if (i == index) {
-    c_refmv.as_int = 0;
-    addmv_and_shuffle(candidate_mvs, candidate_scores,
-                      &index, c_refmv, candidate_scores[3]+1 );
+  if (i == MAX_MV_REF_CANDIDATES) {
+    candidate_mvs[MAX_MV_REF_CANDIDATES-1].as_int = 0;
   }
 
   // Copy over the candidate list.

vp9/encoder/vp9_block.h

@@ -70,7 +70,7 @@ typedef struct {
   PARTITION_INFO partition_info;
   int_mv best_ref_mv;
   int_mv second_best_ref_mv;
-  int_mv ref_mvs[MAX_REF_FRAMES][MAX_MV_REFS];
+  int_mv ref_mvs[MAX_REF_FRAMES][MAX_MV_REF_CANDIDATES];
   int rate;
   int distortion;
   int64_t intra_error;

vp9/encoder/vp9_encodeframe.c

@@ -395,8 +395,7 @@ static unsigned int pick_best_mv_ref(MACROBLOCK *x,
          vp9_mv_bit_cost(&target_mv, &mv_ref_list[0], x->nmvjointcost,
                          x->mvcost, 96, xd->allow_high_precision_mv);
 
-  // Use 4 for now : for (i = 1; i < MAX_MV_REFS; ++i ) {
-  for (i = 1; i < 4; ++i) {
+  for (i = 1; i < MAX_MV_REF_CANDIDATES; ++i) {
     // If we see a 0,0 reference vector for a second time we have reached
     // the end of the list of valid candidate vectors.
     if (!mv_ref_list[i].as_int) {

vp9/encoder/vp9_onyx_if.c

@@ -3821,7 +3821,7 @@ static void encode_frame_to_data_rate
   {
     FILE *f = fopen("mv_ref_dist.stt", "a");
     unsigned int i;
-    for (i = 0; i < MAX_MV_REFS; ++i) {
+    for (i = 0; i < MAX_MV_REF_CANDIDATES; ++i) {
       fprintf(f, "%10d", cpi->best_ref_index_counts[0][i]);
     }
     fprintf(f, "\n" );

vp9/encoder/vp9_onyx_int.h

@@ -790,7 +790,7 @@ typedef struct VP9_COMP {
   unsigned int switchable_interp_count[VP9_SWITCHABLE_FILTERS + 1]
                                       [VP9_SWITCHABLE_FILTERS];
 #if CONFIG_NEW_MVREF
-  unsigned int best_ref_index_counts[MAX_REF_FRAMES][MAX_MV_REFS];
+  unsigned int best_ref_index_counts[MAX_REF_FRAMES][MAX_MV_REF_CANDIDATES];
 #endif
 
 } VP9_COMP;