libvpx: Pull from upstream

source/libvpx/vp9/encoder/vp9_encoder.c

Index: source/libvpx/vp9/encoder/vp9_encoder.c
diff --git a/source/libvpx/vp9/encoder/vp9_encoder.c b/source/libvpx/vp9/encoder/vp9_encoder.c
index 3c0f52e145a2660d0bc22a6c9dde79bbda8aa265..b79bc00d23720b0a33fd2b6f156a102344c7a220 100644
--- a/source/libvpx/vp9/encoder/vp9_encoder.c
+++ b/source/libvpx/vp9/encoder/vp9_encoder.c
@@ -756,6 +756,8 @@ static void init_config(struct VP9_COMP *cpi, VP9EncoderConfig *oxcf) {
   cm->height = oxcf->height;
   vp9_alloc_compressor_data(cpi);
 
+  cpi->svc.temporal_layering_mode = oxcf->temporal_layering_mode;
+
   // Single thread case: use counts in common.
   cpi->td.counts = &cm->counts;
 
@@ -2054,6 +2056,65 @@ void vp9_remove_compressor(VP9_COMP *cpi) {
 #endif
 }
 
+/* TODO(yaowu): The block_variance calls the unoptimized versions of variance()
+ * and highbd_8_variance(). It should not.
+ */
+static void encoder_variance(const uint8_t *a, int  a_stride,
+                             const uint8_t *b, int  b_stride,
+                             int  w, int  h, unsigned int *sse, int *sum) {
+  int i, j;
+
+  *sum = 0;
+  *sse = 0;
+
+  for (i = 0; i < h; i++) {
+    for (j = 0; j < w; j++) {
+      const int diff = a[j] - b[j];
+      *sum += diff;
+      *sse += diff * diff;
+    }
+
+    a += a_stride;
+    b += b_stride;
+  }
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static void encoder_highbd_variance64(const uint8_t *a8, int  a_stride,
+                                      const uint8_t *b8, int  b_stride,
+                                      int w, int h, uint64_t *sse,
+                                      uint64_t *sum) {
+  int i, j;
+
+  uint16_t *a = CONVERT_TO_SHORTPTR(a8);
+  uint16_t *b = CONVERT_TO_SHORTPTR(b8);
+  *sum = 0;
+  *sse = 0;
+
+  for (i = 0; i < h; i++) {
+    for (j = 0; j < w; j++) {
+      const int diff = a[j] - b[j];
+      *sum += diff;
+      *sse += diff * diff;
+    }
+    a += a_stride;
+    b += b_stride;
+  }
+}
+
+static void encoder_highbd_8_variance(const uint8_t *a8, int  a_stride,
+                                      const uint8_t *b8, int  b_stride,
+                                      int w, int h,
+                                      unsigned int *sse, int *sum) {
+  uint64_t sse_long = 0;
+  uint64_t sum_long = 0;
+  encoder_highbd_variance64(a8, a_stride, b8, b_stride, w, h,
+                            &sse_long, &sum_long);
+  *sse = (unsigned int)sse_long;
+  *sum = (int)sum_long;
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static int64_t get_sse(const uint8_t *a, int a_stride,
                        const uint8_t *b, int b_stride,
                        int width, int height) {
@@ -2065,15 +2126,15 @@ static int64_t get_sse(const uint8_t *a, int a_stride,
   int x, y;
 
   if (dw > 0) {
-    variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
-             dw, height, &sse, &sum);
+    encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
+                     dw, height, &sse, &sum);
     total_sse += sse;
   }
 
   if (dh > 0) {
-    variance(&a[(height - dh) * a_stride], a_stride,
-             &b[(height - dh) * b_stride], b_stride,
-             width - dw, dh, &sse, &sum);
+    encoder_variance(&a[(height - dh) * a_stride], a_stride,
+                     &b[(height - dh) * b_stride], b_stride,
+                     width - dw, dh, &sse, &sum);
     total_sse += sse;
   }
 
@@ -2126,14 +2187,15 @@ static int64_t highbd_get_sse(const uint8_t *a, int a_stride,
   unsigned int sse = 0;
   int sum = 0;
   if (dw > 0) {
-    highbd_8_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
-                      dw, height, &sse, &sum);
+    encoder_highbd_8_variance(&a[width - dw], a_stride,
+                              &b[width - dw], b_stride,
+                              dw, height, &sse, &sum);
     total_sse += sse;
   }
   if (dh > 0) {
-    highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
-                      &b[(height - dh) * b_stride], b_stride,
-                      width - dw, dh, &sse, &sum);
+    encoder_highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
+                              &b[(height - dh) * b_stride], b_stride,
+                              width - dw, dh, &sse, &sum);
     total_sse += sse;
   }
   for (y = 0; y < height / 16; ++y) {
@@ -2265,8 +2327,9 @@ static void generate_psnr_packet(VP9_COMP *cpi) {
     pkt.data.psnr.psnr[i] = psnr.psnr[i];
   }
   pkt.kind = VPX_CODEC_PSNR_PKT;
-  if (is_two_pass_svc(cpi))
-    cpi->svc.layer_context[cpi->svc.spatial_layer_id].psnr_pkt = pkt.data.psnr;
+  if (cpi->use_svc)
+    cpi->svc.layer_context[cpi->svc.spatial_layer_id *
+        cpi->svc.number_temporal_layers].psnr_pkt = pkt.data.psnr;
   else
     vpx_codec_pkt_list_add(cpi->output_pkt_list, &pkt);
 }
@@ -3667,9 +3730,11 @@ static void encode_frame_to_data_rate(VP9_COMP *cpi,
   }
   cm->prev_frame = cm->cur_frame;
 
-  if (is_two_pass_svc(cpi))
-    cpi->svc.layer_context[cpi->svc.spatial_layer_id].last_frame_type =
-        cm->frame_type;
+  if (cpi->use_svc)
+    cpi->svc.layer_context[cpi->svc.spatial_layer_id *
+                           cpi->svc.number_temporal_layers +
+                           cpi->svc.temporal_layer_id].last_frame_type =
+                               cm->frame_type;
 }
 
 static void SvcEncode(VP9_COMP *cpi, size_t *size, uint8_t *dest,
@@ -3930,6 +3995,8 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
 #endif
     if (oxcf->pass == 2)
       vp9_restore_layer_context(cpi);
+  } else if (is_one_pass_cbr_svc(cpi)) {
+    vp9_one_pass_cbr_svc_start_layer(cpi);
   }
 
   vpx_usec_timer_start(&cmptimer);
@@ -3948,9 +4015,11 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
   // Normal defaults
   cm->reset_frame_context = 0;
   cm->refresh_frame_context = 1;
-  cpi->refresh_last_frame = 1;
-  cpi->refresh_golden_frame = 0;
-  cpi->refresh_alt_ref_frame = 0;
+  if (!is_one_pass_cbr_svc(cpi)) {
+    cpi->refresh_last_frame = 1;
+    cpi->refresh_golden_frame = 0;
+    cpi->refresh_alt_ref_frame = 0;
+  }
 
   // Should we encode an arf frame.
   arf_src_index = get_arf_src_index(cpi);
@@ -4006,12 +4075,11 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
     }
 
     // Read in the source frame.
-#if CONFIG_SPATIAL_SVC
-    if (is_two_pass_svc(cpi))
+    if (cpi->use_svc)
       source = vp9_svc_lookahead_pop(cpi, cpi->lookahead, flush);
     else
-#endif
       source = vp9_lookahead_pop(cpi->lookahead, flush);
+
     if (source != NULL) {
       cm->show_frame = 1;
       cm->intra_only = 0;
@@ -4060,8 +4128,7 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
     adjust_frame_rate(cpi, source);
   }
 
-  if (cpi->svc.number_temporal_layers > 1 &&
-      oxcf->rc_mode == VPX_CBR) {
+  if (is_one_pass_cbr_svc(cpi)) {
     vp9_update_temporal_layer_framerate(cpi);
     vp9_restore_layer_context(cpi);
   }
@@ -4143,11 +4210,10 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
   }
 
   // Save layer specific state.
-  if ((cpi->svc.number_temporal_layers > 1 &&
-       oxcf->rc_mode == VPX_CBR) ||
-      ((cpi->svc.number_temporal_layers > 1 ||
-        cpi->svc.number_spatial_layers > 1) &&
-       oxcf->pass == 2)) {
+  if (is_one_pass_cbr_svc(cpi) ||
+        ((cpi->svc.number_temporal_layers > 1 ||
+          cpi->svc.number_spatial_layers > 1) &&
+         oxcf->pass == 2)) {
     vp9_save_layer_context(cpi);
   }
 
@@ -4343,6 +4409,12 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
       // May need the empty frame after an visible frame.
       cpi->svc.encode_empty_frame_state = NEED_TO_ENCODE;
     }
+  } else if (is_one_pass_cbr_svc(cpi)) {
+    if (cm->show_frame) {
+      ++cpi->svc.spatial_layer_to_encode;
+      if (cpi->svc.spatial_layer_to_encode >= cpi->svc.number_spatial_layers)
+        cpi->svc.spatial_layer_to_encode = 0;
+    }
   }
   return 0;
 }