Move 4x4 chroma tu info to bottom right cu when intra is split into 4x4

src/encode_coding_tree.c

@@ -352,8 +352,8 @@ void kvz_encode_last_significant_xy(cabac_data_t * const cabac,
 }
 
 static void encode_chroma_tu(encoder_state_t* const state, int x, int y, int depth, const uint8_t width_c, const cu_info_t* cur_pu, int8_t* scan_idx, lcu_coeff_t* coeff, uint8_t joint_chroma) {
-  int x_local = (x >> 1) % LCU_WIDTH_C;
+  int x_local = ((x & ~7) >> 1) % LCU_WIDTH_C;
-  int y_local = (y >> 1) % LCU_WIDTH_C;
+  int y_local = ((y & ~7) >> 1) % LCU_WIDTH_C;
   cabac_data_t* const cabac = &state->cabac;
   *scan_idx = kvz_get_scan_order(cur_pu->type, cur_pu->intra.mode_chroma, depth);
   if(!joint_chroma){
@@ -367,7 +367,7 @@ static void encode_chroma_tu(encoder_state_t* const state, int x, int y, int dep
         // TODO: transform skip for chroma blocks
         CABAC_BIN(cabac, 0, "transform_skip_flag");
       }
-      kvz_encode_coeff_nxn(state, &state->cabac, coeff_u, width_c, 1, *scan_idx, NULL, false);
+      kvz_encode_coeff_nxn(state, &state->cabac, coeff_u, width_c, COLOR_U, *scan_idx, NULL, false);
     }
 
     if (cbf_is_set(cur_pu->cbf, depth, COLOR_V)) {
@@ -375,7 +375,7 @@ static void encode_chroma_tu(encoder_state_t* const state, int x, int y, int dep
         cabac->cur_ctx = &cabac->ctx.transform_skip_model_chroma;
         CABAC_BIN(cabac, 0, "transform_skip_flag");
       }
-      kvz_encode_coeff_nxn(state, &state->cabac, coeff_v, width_c, 2, *scan_idx, NULL, false);
+      kvz_encode_coeff_nxn(state, &state->cabac, coeff_v, width_c, COLOR_V, *scan_idx, NULL, false);
     }
   }
   else {
@@ -384,7 +384,7 @@ static void encode_chroma_tu(encoder_state_t* const state, int x, int y, int dep
       cabac->cur_ctx = &cabac->ctx.transform_skip_model_chroma;
       CABAC_BIN(cabac, 0, "transform_skip_flag");
     }
-    kvz_encode_coeff_nxn(state, &state->cabac, coeff_uv, width_c, 2, *scan_idx, NULL, false);
+    kvz_encode_coeff_nxn(state, &state->cabac, coeff_uv, width_c, COLOR_V, *scan_idx, NULL, false);
     
   }
 }
@@ -444,8 +444,6 @@ static void encode_transform_unit(encoder_state_t * const state,
     } else {
       // Time to to code the chroma transform blocks. Move to the top-left
       // corner of the block.
-      x -= 4;
-      y -= 4;
       cur_pu = kvz_cu_array_at_const((const cu_array_t *)frame->cu_array, x, y);
     }
   }
@@ -485,7 +483,7 @@ static void encode_transform_coeff(encoder_state_t * const state,
   // containing CU.
   const int x_cu = 8 * (x / 8);
   const int y_cu = 8 * (y / 8);
-  const cu_info_t *cur_cu = kvz_cu_array_at_const(frame->cu_array, x_cu, y_cu);
+  const cu_info_t *cur_cu = kvz_cu_array_at_const(frame->cu_array, x, y);
 
   // NxN signifies implicit transform split at the first transform level.
   // There is a similar implicit split for inter, but it is only used when

src/intra.c

@@ -1535,6 +1535,11 @@ void kvz_intra_recon_cu(
   const int8_t mode_luma = search_data->pred_cu.intra.mode;
   const int8_t mode_chroma= search_data->pred_cu.intra.mode_chroma;
 
+  if(mode_chroma != -1 && mode_luma == -1) {
+    x &= ~7;
+    y &= ~7;
+  }
+  
   if (mode_luma != -1 && mode_chroma != -1) {
     if (search_data->pred_cu.intra.mip_flag) {
       assert(mode_luma == mode_chroma && "Chroma mode must be derived from luma mode if block uses MIP.");

src/search.c

@@ -608,7 +608,7 @@ void kvz_select_jccr_mode(
 {
   const vector2d_t lcu_px = { (SUB_SCU(x_px) & ~7) / 2, (SUB_SCU(y_px) & ~7) / 2 };
   const int width = (depth < MAX_DEPTH) ? LCU_WIDTH >> (depth + 1) : LCU_WIDTH >> depth;
-  if (pred_cu == NULL) pred_cu = LCU_GET_CU_AT_PX(lcu, lcu_px.x * 2, lcu_px.y * 2);
+  if (pred_cu == NULL) pred_cu = LCU_GET_CU_AT_PX(lcu, SUB_SCU(x_px), SUB_SCU(y_px));
   assert(pred_cu->depth == pred_cu->tr_depth && "jccr does not support transform splitting");
   if (cost_out == NULL && pred_cu->joint_cb_cr == 0) {
     return;
@@ -650,23 +650,23 @@ void kvz_select_jccr_mode(
     cbf_mask = pred_cu->joint_cb_cr - 1;
     CABAC_FBITS_UPDATE(cabac, &(cabac->ctx.joint_cb_cr[cbf_mask]), 1, joint_cbcr_tr_tree_bits, "jccr_flag");
   }
-  int ssd = 0;
+  unsigned ssd = 0;
-  int joint_ssd = 0;
+  unsigned joint_ssd = 0;
   if (!state->encoder_control->cfg.lossless) {
-    int index = lcu_px.y * LCU_WIDTH_C + lcu_px.x;
+    const int index = lcu_px.y * LCU_WIDTH_C + lcu_px.x;
-    int ssd_u = kvz_pixels_calc_ssd(&lcu->ref.u[index], &lcu->rec.u[index],
+    const unsigned ssd_u = kvz_pixels_calc_ssd(&lcu->ref.u[index], &lcu->rec.u[index],
       LCU_WIDTH_C, LCU_WIDTH_C,
       width);
-    int ssd_v = kvz_pixels_calc_ssd(&lcu->ref.v[index], &lcu->rec.v[index],
+    const unsigned ssd_v = kvz_pixels_calc_ssd(&lcu->ref.v[index], &lcu->rec.v[index],
       LCU_WIDTH_C, LCU_WIDTH_C,
       width);
     ssd = ssd_u + ssd_v;
 
     if (pred_cu->joint_cb_cr) {
-      int ssd_u_joint = kvz_pixels_calc_ssd(&lcu->ref.u[index], &lcu->rec.joint_u[index],
+      const unsigned ssd_u_joint = kvz_pixels_calc_ssd(&lcu->ref.u[index], &lcu->rec.joint_u[index],
         LCU_WIDTH_C, LCU_WIDTH_C,
         width);
-      int ssd_v_joint = kvz_pixels_calc_ssd(&lcu->ref.v[index], &lcu->rec.joint_v[index],
+      const unsigned ssd_v_joint = kvz_pixels_calc_ssd(&lcu->ref.v[index], &lcu->rec.joint_v[index],
         LCU_WIDTH_C, LCU_WIDTH_C,
         width);
       joint_ssd = ssd_u_joint + ssd_v_joint;      
@@ -1009,13 +1009,13 @@ static double search_cu(encoder_state_t * const state, int x, int y, int depth,
         }
         intra_search.pred_cu.intra.mode = -1; // skip luma
         kvz_intra_recon_cu(state,
-                           x & ~7, y & ~7, // TODO: as does this
+                           x, y, // TODO: as does this
                            depth, &intra_search,
                            NULL,
                            lcu);
         if(depth != 0 && state->encoder_control->cfg.jccr && ctrl->cfg.rdo < 3) {
           kvz_select_jccr_mode(state,
-                               x & ~7, y & ~7,
+                               x, y,
                                depth,
                                NULL,
                                lcu,
@@ -1074,7 +1074,7 @@ static double search_cu(encoder_state_t * const state, int x, int y, int depth,
                                   false);
         if (cur_cu->depth == cur_cu->tr_depth && state->encoder_control->cfg.jccr && cur_cu->joint_cb_cr) {
           kvz_select_jccr_mode(state,
-            x & ~7, y & ~7,
+            x, y,
             depth,
             NULL,
             lcu,

src/search_intra.c

@@ -367,7 +367,7 @@ static double search_intra_trdepth(
       pred_cu->intra.mode_chroma = chroma_mode;
       pred_cu->joint_cb_cr= 4; // TODO: Maybe check the jccr mode here also but holy shit is the interface of search_intra_rdo bad currently
       kvz_intra_recon_cu(state,
-                         x_px & ~7, y_px & ~7,
+                         x_px, y_px,
                          depth, search_data,
                          pred_cu, 
                          lcu);