Make encoder and decoder share common thread function

Moved vp9_accumulate_frame_counts to vp9_thread_common.c to
eliminate the duplicate code.

Change-Id: I9cf506d729603c8bf1494b4c86a3b7d47af1917a

diff --git a/vp9/common/vp9_entropymode.h b/vp9/common/vp9_entropymode.h
index 972509d..f4e20e1 100644 (file)
--- a/vp9/common/vp9_entropymode.h
+++ b/vp9/common/vp9_entropymode.h
@@ -54,7 +54,7 @@ typedef struct frame_contexts {
   int initialized;
 } FRAME_CONTEXT;
 
-typedef struct {
+typedef struct FRAME_COUNTS {
   unsigned int y_mode[BLOCK_SIZE_GROUPS][INTRA_MODES];
   unsigned int uv_mode[INTRA_MODES][INTRA_MODES];
   unsigned int partition[PARTITION_CONTEXTS][PARTITION_TYPES];

diff --git a/vp9/common/vp9_thread_common.c b/vp9/common/vp9_thread_common.c
index 98799a3..1a93a34 100644 (file)
--- a/vp9/common/vp9_thread_common.c
+++ b/vp9/common/vp9_thread_common.c
@@ -10,6 +10,7 @@
 
 #include "./vpx_config.h"
 #include "vpx_mem/vpx_mem.h"
+#include "vp9/common/vp9_entropymode.h"
 #include "vp9/common/vp9_thread_common.h"
 #include "vp9/common/vp9_reconinter.h"
 
@@ -299,3 +300,119 @@ void vp9_loop_filter_dealloc(VP9LfSync *lf_sync) {
     vp9_zero(*lf_sync);
   }
 }
+
+// Accumulate frame counts.
+void vp9_accumulate_frame_counts(VP9_COMMON *cm, FRAME_COUNTS *counts,
+                                 int is_dec) {
+  int i, j, k, l, m;
+
+  for (i = 0; i < BLOCK_SIZE_GROUPS; i++)
+    for (j = 0; j < INTRA_MODES; j++)
+      cm->counts.y_mode[i][j] += counts->y_mode[i][j];
+
+  for (i = 0; i < INTRA_MODES; i++)
+    for (j = 0; j < INTRA_MODES; j++)
+      cm->counts.uv_mode[i][j] += counts->uv_mode[i][j];
+
+  for (i = 0; i < PARTITION_CONTEXTS; i++)
+    for (j = 0; j < PARTITION_TYPES; j++)
+      cm->counts.partition[i][j] += counts->partition[i][j];
+
+  if (is_dec) {
+    int n;
+    for (i = 0; i < TX_SIZES; i++)
+      for (j = 0; j < PLANE_TYPES; j++)
+        for (k = 0; k < REF_TYPES; k++)
+          for (l = 0; l < COEF_BANDS; l++)
+            for (m = 0; m < COEFF_CONTEXTS; m++) {
+              cm->counts.eob_branch[i][j][k][l][m] +=
+                  counts->eob_branch[i][j][k][l][m];
+              for (n = 0; n < UNCONSTRAINED_NODES + 1; n++)
+                cm->counts.coef[i][j][k][l][m][n] +=
+                    counts->coef[i][j][k][l][m][n];
+            }
+  } else {
+    for (i = 0; i < TX_SIZES; i++)
+      for (j = 0; j < PLANE_TYPES; j++)
+        for (k = 0; k < REF_TYPES; k++)
+          for (l = 0; l < COEF_BANDS; l++)
+            for (m = 0; m < COEFF_CONTEXTS; m++)
+              cm->counts.eob_branch[i][j][k][l][m] +=
+                  counts->eob_branch[i][j][k][l][m];
+                // In the encoder, cm->counts.coef is only updated at frame
+                // level, so not need to accumulate it here.
+                // for (n = 0; n < UNCONSTRAINED_NODES + 1; n++)
+                //   cm->counts.coef[i][j][k][l][m][n] +=
+                //       counts->coef[i][j][k][l][m][n];
+  }
+
+  for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; i++)
+    for (j = 0; j < SWITCHABLE_FILTERS; j++)
+      cm->counts.switchable_interp[i][j] += counts->switchable_interp[i][j];
+
+  for (i = 0; i < INTER_MODE_CONTEXTS; i++)
+    for (j = 0; j < INTER_MODES; j++)
+      cm->counts.inter_mode[i][j] += counts->inter_mode[i][j];
+
+  for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
+    for (j = 0; j < 2; j++)
+      cm->counts.intra_inter[i][j] += counts->intra_inter[i][j];
+
+  for (i = 0; i < COMP_INTER_CONTEXTS; i++)
+    for (j = 0; j < 2; j++)
+      cm->counts.comp_inter[i][j] += counts->comp_inter[i][j];
+
+  for (i = 0; i < REF_CONTEXTS; i++)
+    for (j = 0; j < 2; j++)
+      for (k = 0; k < 2; k++)
+      cm->counts.single_ref[i][j][k] += counts->single_ref[i][j][k];
+
+  for (i = 0; i < REF_CONTEXTS; i++)
+    for (j = 0; j < 2; j++)
+      cm->counts.comp_ref[i][j] += counts->comp_ref[i][j];
+
+  for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
+    for (j = 0; j < TX_SIZES; j++)
+      cm->counts.tx.p32x32[i][j] += counts->tx.p32x32[i][j];
+
+    for (j = 0; j < TX_SIZES - 1; j++)
+      cm->counts.tx.p16x16[i][j] += counts->tx.p16x16[i][j];
+
+    for (j = 0; j < TX_SIZES - 2; j++)
+      cm->counts.tx.p8x8[i][j] += counts->tx.p8x8[i][j];
+  }
+
+  for (i = 0; i < SKIP_CONTEXTS; i++)
+    for (j = 0; j < 2; j++)
+      cm->counts.skip[i][j] += counts->skip[i][j];
+
+  for (i = 0; i < MV_JOINTS; i++)
+    cm->counts.mv.joints[i] += counts->mv.joints[i];
+
+  for (k = 0; k < 2; k++) {
+    nmv_component_counts *comps = &cm->counts.mv.comps[k];
+    nmv_component_counts *comps_t = &counts->mv.comps[k];
+
+    for (i = 0; i < 2; i++) {
+      comps->sign[i] += comps_t->sign[i];
+      comps->class0_hp[i] += comps_t->class0_hp[i];
+      comps->hp[i] += comps_t->hp[i];
+    }
+
+    for (i = 0; i < MV_CLASSES; i++)
+      comps->classes[i] += comps_t->classes[i];
+
+    for (i = 0; i < CLASS0_SIZE; i++) {
+      comps->class0[i] += comps_t->class0[i];
+      for (j = 0; j < MV_FP_SIZE; j++)
+        comps->class0_fp[i][j] += comps_t->class0_fp[i][j];
+    }
+
+    for (i = 0; i < MV_OFFSET_BITS; i++)
+      for (j = 0; j < 2; j++)
+        comps->bits[i][j] += comps_t->bits[i][j];
+
+    for (i = 0; i < MV_FP_SIZE; i++)
+      comps->fp[i] += comps_t->fp[i];
+  }
+}

diff --git a/vp9/common/vp9_thread_common.h b/vp9/common/vp9_thread_common.h
index bca357e..3b3a699 100644 (file)
--- a/vp9/common/vp9_thread_common.h
+++ b/vp9/common/vp9_thread_common.h
@@ -15,6 +15,7 @@
 #include "vp9/common/vp9_thread.h"
 
 struct VP9Common;
+struct FRAME_COUNTS;
 
 // Loopfilter row synchronization
 typedef struct VP9LfSyncData {
@@ -50,4 +51,7 @@ void vp9_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame,
                               VP9Worker *workers, int num_workers,
                               VP9LfSync *lf_sync);
 
+void vp9_accumulate_frame_counts(struct VP9Common *cm,
+                                 struct FRAME_COUNTS *counts, int is_dec);
+
 #endif  // VP9_COMMON_VP9_LOOPFILTER_THREAD_H_

diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index 87aa161..1d0d8f6 100644 (file)
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -1085,105 +1085,6 @@ static int compare_tile_buffers(const void *a, const void *b) {
   }
 }
 
-// Accumulate frame counts.
-static void accumulate_frame_counts(VP9_COMMON *cm, FRAME_COUNTS *counts) {
-  int i, j, k, l, m, n;
-
-  for (i = 0; i < BLOCK_SIZE_GROUPS; i++)
-    for (j = 0; j < INTRA_MODES; j++)
-      cm->counts.y_mode[i][j] += counts->y_mode[i][j];
-
-  for (i = 0; i < INTRA_MODES; i++)
-    for (j = 0; j < INTRA_MODES; j++)
-      cm->counts.uv_mode[i][j] += counts->uv_mode[i][j];
-
-  for (i = 0; i < PARTITION_CONTEXTS; i++)
-    for (j = 0; j < PARTITION_TYPES; j++)
-      cm->counts.partition[i][j] += counts->partition[i][j];
-
-  for (i = 0; i < TX_SIZES; i++)
-    for (j = 0; j < PLANE_TYPES; j++)
-      for (k = 0; k < REF_TYPES; k++)
-        for (l = 0; l < COEF_BANDS; l++)
-          for (m = 0; m < COEFF_CONTEXTS; m++) {
-            cm->counts.eob_branch[i][j][k][l][m] +=
-                counts->eob_branch[i][j][k][l][m];
-            for (n = 0; n < UNCONSTRAINED_NODES + 1; n++)
-              cm->counts.coef[i][j][k][l][m][n] +=
-                  counts->coef[i][j][k][l][m][n];
-          }
-
-  for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; i++)
-    for (j = 0; j < SWITCHABLE_FILTERS; j++)
-      cm->counts.switchable_interp[i][j] += counts->switchable_interp[i][j];
-
-  for (i = 0; i < INTER_MODE_CONTEXTS; i++)
-    for (j = 0; j < INTER_MODES; j++)
-      cm->counts.inter_mode[i][j] += counts->inter_mode[i][j];
-
-  for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
-    for (j = 0; j < 2; j++)
-      cm->counts.intra_inter[i][j] += counts->intra_inter[i][j];
-
-  for (i = 0; i < COMP_INTER_CONTEXTS; i++)
-    for (j = 0; j < 2; j++)
-      cm->counts.comp_inter[i][j] += counts->comp_inter[i][j];
-
-  for (i = 0; i < REF_CONTEXTS; i++)
-    for (j = 0; j < 2; j++)
-      for (k = 0; k < 2; k++)
-      cm->counts.single_ref[i][j][k] += counts->single_ref[i][j][k];
-
-  for (i = 0; i < REF_CONTEXTS; i++)
-    for (j = 0; j < 2; j++)
-      cm->counts.comp_ref[i][j] += counts->comp_ref[i][j];
-
-  for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
-    for (j = 0; j < TX_SIZES; j++)
-      cm->counts.tx.p32x32[i][j] += counts->tx.p32x32[i][j];
-
-    for (j = 0; j < TX_SIZES - 1; j++)
-      cm->counts.tx.p16x16[i][j] += counts->tx.p16x16[i][j];
-
-    for (j = 0; j < TX_SIZES - 2; j++)
-      cm->counts.tx.p8x8[i][j] += counts->tx.p8x8[i][j];
-  }
-
-  for (i = 0; i < SKIP_CONTEXTS; i++)
-    for (j = 0; j < 2; j++)
-      cm->counts.skip[i][j] += counts->skip[i][j];
-
-  for (i = 0; i < MV_JOINTS; i++)
-    cm->counts.mv.joints[i] += counts->mv.joints[i];
-
-  for (k = 0; k < 2; k++) {
-    nmv_component_counts *comps = &cm->counts.mv.comps[k];
-    nmv_component_counts *comps_t = &counts->mv.comps[k];
-
-    for (i = 0; i < 2; i++) {
-      comps->sign[i] += comps_t->sign[i];
-      comps->class0_hp[i] += comps_t->class0_hp[i];
-      comps->hp[i] += comps_t->hp[i];
-    }
-
-    for (i = 0; i < MV_CLASSES; i++)
-      comps->classes[i] += comps_t->classes[i];
-
-    for (i = 0; i < CLASS0_SIZE; i++) {
-      comps->class0[i] += comps_t->class0[i];
-      for (j = 0; j < MV_FP_SIZE; j++)
-        comps->class0_fp[i][j] += comps_t->class0_fp[i][j];
-    }
-
-    for (i = 0; i < MV_OFFSET_BITS; i++)
-      for (j = 0; j < 2; j++)
-        comps->bits[i][j] += comps_t->bits[i][j];
-
-    for (i = 0; i < MV_FP_SIZE; i++)
-      comps->fp[i] += comps_t->fp[i];
-  }
-}
-
 static const uint8_t *decode_tiles_mt(VP9Decoder *pbi,
                                       const uint8_t *data,
                                       const uint8_t *data_end) {
@@ -1333,7 +1234,7 @@ static const uint8_t *decode_tiles_mt(VP9Decoder *pbi,
       for (i = 0; i < num_workers; ++i) {
         TileWorkerData *const tile_data =
             (TileWorkerData*)pbi->tile_workers[i].data1;
-        accumulate_frame_counts(cm, &tile_data->counts);
+        vp9_accumulate_frame_counts(cm, &tile_data->counts, 1);
       }
     }
   }

diff --git a/vp9/encoder/vp9_ethread.c b/vp9/encoder/vp9_ethread.c
index 12fb4d1..9916cc0 100644 (file)
--- a/vp9/encoder/vp9_ethread.c
+++ b/vp9/encoder/vp9_ethread.c
@@ -12,105 +12,6 @@
 #include "vp9/encoder/vp9_encoder.h"
 #include "vp9/encoder/vp9_ethread.h"
 
-static void accumulate_frame_counts(VP9_COMMON *cm, ThreadData *td) {
-  int i, j, k, l, m;
-
-  for (i = 0; i < BLOCK_SIZE_GROUPS; i++)
-    for (j = 0; j < INTRA_MODES; j++)
-      cm->counts.y_mode[i][j] += td->counts->y_mode[i][j];
-
-  for (i = 0; i < INTRA_MODES; i++)
-    for (j = 0; j < INTRA_MODES; j++)
-      cm->counts.uv_mode[i][j] += td->counts->uv_mode[i][j];
-
-  for (i = 0; i < PARTITION_CONTEXTS; i++)
-    for (j = 0; j < PARTITION_TYPES; j++)
-      cm->counts.partition[i][j] += td->counts->partition[i][j];
-
-  for (i = 0; i < TX_SIZES; i++)
-    for (j = 0; j < PLANE_TYPES; j++)
-      for (k = 0; k < REF_TYPES; k++)
-        for (l = 0; l < COEF_BANDS; l++)
-          for (m = 0; m < COEFF_CONTEXTS; m++)
-            cm->counts.eob_branch[i][j][k][l][m] +=
-                td->counts->eob_branch[i][j][k][l][m];
-              // cm->counts.coef is only updated at frame level, so not need
-              // to accumulate it here.
-              // for (n = 0; n < UNCONSTRAINED_NODES + 1; n++)
-              //   cm->counts.coef[i][j][k][l][m][n] +=
-              //       td->counts->coef[i][j][k][l][m][n];
-
-  for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; i++)
-    for (j = 0; j < SWITCHABLE_FILTERS; j++)
-      cm->counts.switchable_interp[i][j] += td->counts->switchable_interp[i][j];
-
-  for (i = 0; i < INTER_MODE_CONTEXTS; i++)
-    for (j = 0; j < INTER_MODES; j++)
-      cm->counts.inter_mode[i][j] += td->counts->inter_mode[i][j];
-
-  for (i = 0; i < INTRA_INTER_CONTEXTS; i++)
-    for (j = 0; j < 2; j++)
-      cm->counts.intra_inter[i][j] += td->counts->intra_inter[i][j];
-
-  for (i = 0; i < COMP_INTER_CONTEXTS; i++)
-    for (j = 0; j < 2; j++)
-      cm->counts.comp_inter[i][j] += td->counts->comp_inter[i][j];
-
-  for (i = 0; i < REF_CONTEXTS; i++)
-    for (j = 0; j < 2; j++)
-      for (k = 0; k < 2; k++)
-      cm->counts.single_ref[i][j][k] += td->counts->single_ref[i][j][k];
-
-  for (i = 0; i < REF_CONTEXTS; i++)
-    for (j = 0; j < 2; j++)
-      cm->counts.comp_ref[i][j] += td->counts->comp_ref[i][j];
-
-  for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
-    for (j = 0; j < TX_SIZES; j++)
-      cm->counts.tx.p32x32[i][j] += td->counts->tx.p32x32[i][j];
-
-    for (j = 0; j < TX_SIZES - 1; j++)
-      cm->counts.tx.p16x16[i][j] += td->counts->tx.p16x16[i][j];
-
-    for (j = 0; j < TX_SIZES - 2; j++)
-      cm->counts.tx.p8x8[i][j] += td->counts->tx.p8x8[i][j];
-  }
-
-  for (i = 0; i < SKIP_CONTEXTS; i++)
-    for (j = 0; j < 2; j++)
-      cm->counts.skip[i][j] += td->counts->skip[i][j];
-
-  for (i = 0; i < MV_JOINTS; i++)
-    cm->counts.mv.joints[i] += td->counts->mv.joints[i];
-
-  for (k = 0; k < 2; k++) {
-    nmv_component_counts *comps = &cm->counts.mv.comps[k];
-    nmv_component_counts *comps_t = &td->counts->mv.comps[k];
-
-    for (i = 0; i < 2; i++) {
-      comps->sign[i] += comps_t->sign[i];
-      comps->class0_hp[i] += comps_t->class0_hp[i];
-      comps->hp[i] += comps_t->hp[i];
-    }
-
-    for (i = 0; i < MV_CLASSES; i++)
-      comps->classes[i] += comps_t->classes[i];
-
-    for (i = 0; i < CLASS0_SIZE; i++) {
-      comps->class0[i] += comps_t->class0[i];
-      for (j = 0; j < MV_FP_SIZE; j++)
-        comps->class0_fp[i][j] += comps_t->class0_fp[i][j];
-    }
-
-    for (i = 0; i < MV_OFFSET_BITS; i++)
-      for (j = 0; j < 2; j++)
-        comps->bits[i][j] += comps_t->bits[i][j];
-
-    for (i = 0; i < MV_FP_SIZE; i++)
-      comps->fp[i] += comps_t->fp[i];
-  }
-}
-
 static void accumulate_rd_opt(ThreadData *td, ThreadData *td_t) {
   int i, j, k, l, m, n;
 
@@ -267,7 +168,7 @@ void vp9_encode_tiles_mt(VP9_COMP *cpi) {
 
     // Accumulate counters.
     if (i < num_workers - 1) {
-      accumulate_frame_counts(&cpi->common, thread_data->td);
+      vp9_accumulate_frame_counts(cm, thread_data->td->counts, 0);
       accumulate_rd_opt(&cpi->td, thread_data->td);
     }
   }