[mrl] Pass multi_ref_idx to generic angular and DC prediction funcs. Fix some outdated comments.

src/intra.c

@@ -212,24 +212,26 @@ static void intra_filter_reference(
 
 
 /**
-* \brief Generage planar prediction.
+* \brief Generate dc prediction.
 * \param log2_width    Log2 of width, range 2..5.
-* \param in_ref_above  Pointer to -1 index of above reference, length=width*2+1.
+* \param ref_top       Pointer to -1 index of above reference, length=width*2+1.
-* \param in_ref_left   Pointer to -1 index of left reference, length=width*2+1.
+* \param ref_left      Pointer to -1 index of left reference, length=width*2+1.
 * \param dst           Buffer of size width*width.
+* \param multi_ref_idx Multi reference line index for use with MRL.
 */
 static void intra_pred_dc(
   const int_fast8_t log2_width,
   const kvz_pixel *const ref_top,
   const kvz_pixel *const ref_left,
-  kvz_pixel *const out_block)
+  kvz_pixel *const out_block,
+  const uint8_t multi_ref_idx)
 {
   int_fast8_t width = 1 << log2_width;
 
   int_fast16_t sum = 0;
   for (int_fast8_t i = 0; i < width; ++i) {
-    sum += ref_top[i + 1];
+    sum += ref_top[i + 1 + multi_ref_idx];
-    sum += ref_left[i + 1];
+    sum += ref_left[i + 1 + multi_ref_idx];
   }
   
   // JVET_K0122
@@ -549,11 +551,15 @@ void kvz_intra_predict(
   int_fast8_t mode,
   color_t color,
   kvz_pixel *dst,
-  bool filter_boundary)
+  bool filter_boundary,
+  const uint8_t multi_ref_idx)
 {
   const int_fast8_t width = 1 << log2_width;
   const kvz_config *cfg = &state->encoder_control->cfg;
 
+  // MRL only for luma
+  uint8_t multi_ref_index = color == COLOR_Y ? multi_ref_idx : 0;
+
   const kvz_intra_ref *used_ref = &refs->ref;
   if (cfg->intra_smoothing_disabled || color != COLOR_Y || mode == 1 || width == 4) {
     // For chroma, DC and 4x4 blocks, always use unfiltered reference.
@@ -586,9 +592,9 @@ void kvz_intra_predict(
   if (mode == 0) {
     kvz_intra_pred_planar(log2_width, used_ref->top, used_ref->left, dst);
   } else if (mode == 1) {
-    intra_pred_dc(log2_width, used_ref->top, used_ref->left, dst);
+    intra_pred_dc(log2_width, used_ref->top, used_ref->left, dst, multi_ref_index);
   } else {
-    kvz_angular_pred(log2_width, mode, color, used_ref->top, used_ref->left, dst);
+    kvz_angular_pred(log2_width, mode, color, used_ref->top, used_ref->left, dst, multi_ref_index);
   }
 
   // pdpc
@@ -608,7 +614,7 @@ void kvz_intra_build_reference_any(
   const vector2d_t *const pic_px,
   const lcu_t *const lcu,
   kvz_intra_references *const refs,
-  const uint8_t mri)
+  const uint8_t multi_ref_idx)
 {
   assert(log2_width >= 2 && log2_width <= 5);
 
@@ -621,7 +627,7 @@ void kvz_intra_build_reference_any(
   const int_fast8_t width = 1 << log2_width;
 
   // Get multiRefIdx from CU under prediction. Do not use MRL if not luma
-  const uint8_t multi_ref_index = !is_chroma ? mri : 0;
+  const uint8_t multi_ref_index = !is_chroma ? multi_ref_idx : 0;
 
   // Convert luma coordinates to chroma coordinates for chroma.
   const vector2d_t lcu_px = {
@@ -737,7 +743,7 @@ void kvz_intra_build_reference_inner(
   const lcu_t *const lcu,
   kvz_intra_references *const refs,
   bool entropy_sync,
-  const uint8_t mri)
+  const uint8_t multi_ref_idx)
 {
   assert(log2_width >= 2 && log2_width <= 5);
 
@@ -749,7 +755,7 @@ void kvz_intra_build_reference_inner(
   const int_fast8_t width = 1 << log2_width;
 
   // Get multiRefIdx from CU under prediction. Do not use MRL if not luma
-  const uint8_t multi_ref_index = !is_chroma ? mri : 0;
+  const uint8_t multi_ref_index = !is_chroma ? multi_ref_idx : 0;
 
   // Convert luma coordinates to chroma coordinates for chroma.
   const vector2d_t lcu_px = {
@@ -915,6 +921,7 @@ static void intra_recon_tb_leaf(
   int x_scu = SUB_SCU(x);
   int y_scu = SUB_SCU(y);
   const vector2d_t lcu_px = {x_scu >> shift, y_scu >> shift };
+  cu_info_t* cur_cu = LCU_GET_CU_AT_PX(lcu, lcu_px.x, lcu_px.y);
 
   kvz_intra_references refs;
   kvz_intra_build_reference(log2width, color, &luma_px, &pic_px, lcu, &refs, cfg->wpp);
@@ -923,7 +930,7 @@ static void intra_recon_tb_leaf(
   int stride = state->tile->frame->source->stride;
   const bool filter_boundary = color == COLOR_Y && !(cfg->lossless && cfg->implicit_rdpcm);
   if(intra_mode < 68) {
-    kvz_intra_predict(state, &refs, log2width, intra_mode, color, pred, filter_boundary);
+    kvz_intra_predict(state, &refs, log2width, intra_mode, color, pred, filter_boundary, cur_cu->intra.multi_ref_idx);
   } else {
     kvz_pixels_blit(&state->tile->frame->cclm_luma_rec[x / 2 + (y * stride) / 4], pred, width, width, stride / 2, width);
     if(cclm_params == NULL) {

src/intra.h

@@ -79,14 +79,14 @@ int8_t kvz_intra_get_dir_luma_predictor(
   const cu_info_t *const above_pu);
 
 /**
-* \brief Generage angular predictions.
+* \brief Build intra prediction reference buffers.
-* \param width    Width in pixels, range 4..32.
+* \param log2_width    Log2 of width, range 2..5.
-* \param color    What color pixels to use.
+* \param color         What color pixels to use.
-* \param luma_px  Luma coordinates of the prediction block.
+* \param luma_px       Luma coordinates of the prediction block.
-* \param pic_px   Picture dimensions in luma pixels.
+* \param pic_px        Picture dimensions in luma pixels.
-* \param lcu      LCU struct.
+* \param lcu           LCU struct.
-* \param out_left_ref  Left reference pixels, index 0 is the top-left.
+* \param refs          Pointer to top and left references.
-* \param out_top_ref   Top reference pixels, index 0 is the top-left.
+* \param entropy_sync  Indicate that top right is not available if WPP is enabled.
 */
 void kvz_intra_build_reference(
   const int_fast8_t log2_width,
@@ -113,7 +113,8 @@ void kvz_intra_predict(
   int_fast8_t mode,
   color_t color,
   kvz_pixel *dst,
-  bool filter_boundary);
+  bool filter_boundary,
+  const uint8_t multi_ref_idx);
 
 void kvz_intra_recon_cu(
   encoder_state_t *const state,

src/search_intra.c

@@ -480,7 +480,7 @@ static void search_intra_chroma_rough(encoder_state_t * const state,
   kvz_pixels_blit(orig_u, orig_block, width, width, origstride, width);
   for (int i = 0; i < 5; ++i) {
     if (modes[i] == -1) continue;
-    kvz_intra_predict(state, refs_u, log2_width_c, modes[i], COLOR_U, pred, false);
+    kvz_intra_predict(state, refs_u, log2_width_c, modes[i], COLOR_U, pred, false, 0);
     //costs[i] += get_cost(encoder_state, pred, orig_block, satd_func, sad_func, width);
     costs[i] += satd_func(pred, orig_block);
   }
@@ -494,7 +494,7 @@ static void search_intra_chroma_rough(encoder_state_t * const state,
   kvz_pixels_blit(orig_v, orig_block, width, width, origstride, width);
   for (int i = 0; i < 5; ++i) {
     if (modes[i] == -1) continue;
-    kvz_intra_predict(state, refs_v, log2_width_c, modes[i], COLOR_V, pred, false);
+    kvz_intra_predict(state, refs_v, log2_width_c, modes[i], COLOR_V, pred, false, 0);
     //costs[i] += get_cost(encoder_state, pred, orig_block, satd_func, sad_func, width);
     costs[i] += satd_func(pred, orig_block);
   }
@@ -588,7 +588,7 @@ static int8_t search_intra_rough(encoder_state_t * const state,
     double costs_out[PARALLEL_BLKS] = { 0 };
     for (int i = 0; i < PARALLEL_BLKS; ++i) {
       if (mode + i * offset <= 66) {
-        kvz_intra_predict(state, refs, log2_width, mode + i * offset, COLOR_Y, preds[i], filter_boundary);
+        kvz_intra_predict(state, refs, log2_width, mode + i * offset, COLOR_Y, preds[i], filter_boundary, 0);
       }
     }
     
@@ -627,7 +627,7 @@ static int8_t search_intra_rough(encoder_state_t * const state,
       if (mode_in_range) {
         for (int i = 0; i < PARALLEL_BLKS; ++i) {
           if (test_modes[i] >= 2 && test_modes[i] <= 66) {
-            kvz_intra_predict(state, refs, log2_width, test_modes[i], COLOR_Y, preds[i], filter_boundary);
+            kvz_intra_predict(state, refs, log2_width, test_modes[i], COLOR_Y, preds[i], filter_boundary, 0);
           }
         }
 
@@ -664,7 +664,7 @@ static int8_t search_intra_rough(encoder_state_t * const state,
     }
 
     if (!has_mode) {
-      kvz_intra_predict(state, refs, log2_width, mode, COLOR_Y, preds[0], filter_boundary);
+      kvz_intra_predict(state, refs, log2_width, mode, COLOR_Y, preds[0], filter_boundary, 0);
       costs[modes_selected] = get_cost(state, preds[0], orig_block, satd_func, sad_func, width);
       modes[modes_selected] = mode;
       ++modes_selected;

src/strategies/avx2/intra-avx2.c

@@ -446,6 +446,25 @@ static void kvz_angular_pred_avx2(
     }
   }
 
+ /**
+ * \brief Generage angular predictions.
+ * \param log2_width    Log2 of width, range 2..5.
+ * \param intra_mode    Angular mode in range 2..34.
+ * \param in_ref_above  Pointer to -1 index of above reference, length=width*2+1.
+ * \param in_ref_left   Pointer to -1 index of left reference, length=width*2+1.
+ * \param dst           Buffer of size width*width.
+ * \param multi_ref_idx Multi reference line index for use with MRL.
+ */
+static void kvz_angular_pred_avx2(
+  const int_fast8_t log2_width,
+  const int_fast8_t intra_mode,
+  const uint8_t *const in_ref_above,
+  const uint8_t *const in_ref_left,
+  uint8_t *const dst,
+  const uint8_t multi_ref_idx)
+{
+  assert(log2_width >= 2 && log2_width <= 5);
+  assert(intra_mode >= 2 && intra_mode <= 34);
   // Flip the block if this is was a horizontal mode.
   if (!vertical_mode) {
     
@@ -459,6 +478,12 @@ static void kvz_angular_pred_avx2(
     const __m128i vseq = _mm_setr_epi32(0, 1, 2, 3);
     const __m128i vidx = _mm_slli_epi32(vseq, log2_width);
 
+  // TODO: implement usage of multi_ref_idx
+
+                                                    // Temporary buffer for modes 11-25.
+                                                    // It only needs to be big enough to hold indices from -width to width-1.
+  uint8_t tmp_ref[2 * 32];
+  const int_fast8_t width = 1 << log2_width;
     // Transpose as 4x4 subblocks
     for (int_fast32_t y = 0; y + 3 < width; y += 4) {
       for (int_fast32_t x = y; x + 3 < width; x += 4) {
@@ -590,7 +615,8 @@ static void kvz_intra_pred_planar_avx2(
 // addends etc can be preinitialized for each position.
 static void pred_filtered_dc_4x4(const uint8_t *ref_top,
                                  const uint8_t *ref_left,
-                                       uint8_t *out_block)
+                                       uint8_t *out_block,
+                                 const uint8_t multi_ref_idx)
 {
   const uint32_t rt_u32 = *(const uint32_t *)(ref_top  + 1);
   const uint32_t rl_u32 = *(const uint32_t *)(ref_left + 1);
@@ -651,7 +677,8 @@ static void pred_filtered_dc_4x4(const uint8_t *ref_top,
 
 static void pred_filtered_dc_8x8(const uint8_t *ref_top,
                                  const uint8_t *ref_left,
-                                       uint8_t *out_block)
+                                       uint8_t *out_block,
+                                 const uint8_t multi_ref_idx)
 {
   const uint64_t rt_u64 = *(const uint64_t *)(ref_top  + 1);
   const uint64_t rl_u64 = *(const uint64_t *)(ref_left + 1);
@@ -755,7 +782,8 @@ static INLINE __m256i cvt_u32_si256(const uint32_t u)
 
 static void pred_filtered_dc_16x16(const uint8_t *ref_top,
                                    const uint8_t *ref_left,
-                                         uint8_t *out_block)
+                                         uint8_t *out_block,
+                                   const uint8_t multi_ref_idx)
 {
   const __m128i rt_128 = _mm_loadu_si128((const __m128i *)(ref_top  + 1));
   const __m128i rl_128 = _mm_loadu_si128((const __m128i *)(ref_left + 1));
@@ -831,7 +859,8 @@ static void pred_filtered_dc_16x16(const uint8_t *ref_top,
 
 static void pred_filtered_dc_32x32(const uint8_t *ref_top,
                                    const uint8_t *ref_left,
-                                         uint8_t *out_block)
+                                         uint8_t *out_block,
+                                   const uint8_t multi_ref_idx)
 {
   const __m256i rt = _mm256_loadu_si256((const __m256i *)(ref_top  + 1));
   const __m256i rl = _mm256_loadu_si256((const __m256i *)(ref_left + 1));
@@ -913,23 +942,26 @@ static void pred_filtered_dc_32x32(const uint8_t *ref_top,
 * \param in_ref_above  Pointer to -1 index of above reference, length=width*2+1.
 * \param in_ref_left   Pointer to -1 index of left reference, length=width*2+1.
 * \param dst           Buffer of size width*width.
+* \param multi_ref_idx Reference line index. May be non-zero when MRL is used.
 */
 static void kvz_intra_pred_filtered_dc_avx2(
   const int_fast8_t log2_width,
   const uint8_t *ref_top,
   const uint8_t *ref_left,
-        uint8_t *out_block)
+        uint8_t *out_block,
+  const uint8_t multi_ref_idx)
 {
   assert(log2_width >= 2 && log2_width <= 5);
 
+  // TODO: implement multi reference index for all subfunctions
   if (log2_width == 2) {
-    pred_filtered_dc_4x4(ref_top, ref_left, out_block);
+    pred_filtered_dc_4x4(ref_top, ref_left, out_block, multi_ref_idx);
   } else if (log2_width == 3) {
-    pred_filtered_dc_8x8(ref_top, ref_left, out_block);
+    pred_filtered_dc_8x8(ref_top, ref_left, out_block, multi_ref_idx);
   } else if (log2_width == 4) {
-    pred_filtered_dc_16x16(ref_top, ref_left, out_block);
+    pred_filtered_dc_16x16(ref_top, ref_left, out_block, multi_ref_idx);
   } else if (log2_width == 5) {
-    pred_filtered_dc_32x32(ref_top, ref_left, out_block);
+    pred_filtered_dc_32x32(ref_top, ref_left, out_block, multi_ref_idx);
   }
 }
 

src/strategies/generic/intra-generic.c

@@ -49,6 +49,7 @@
  * \param in_ref_above  Pointer to -1 index of above reference, length=width*2+1.
  * \param in_ref_left   Pointer to -1 index of left reference, length=width*2+1.
  * \param dst           Buffer of size width*width.
+ * \param multi_ref_idx Multi reference line index for use with MRL.
  */
 static void kvz_angular_pred_generic(
   const int_fast8_t log2_width,
@@ -56,7 +57,8 @@ static void kvz_angular_pred_generic(
   const int_fast8_t channel_type,
   const kvz_pixel *const in_ref_above,
   const kvz_pixel *const in_ref_left,
-  kvz_pixel *const dst)
+  kvz_pixel *const dst,
+  const uint8_t multi_ref_idx)
 {
   
   assert(log2_width >= 2 && log2_width <= 5);
@@ -113,8 +115,7 @@ static void kvz_angular_pred_generic(
 
   uint32_t pred_mode = intra_mode; // ToDo: handle WAIP
 
-  // TODO: pass the multiRefIdx to this function and assign to this variable
+  uint8_t multi_ref_index = multi_ref_idx;
-  uint8_t multi_ref_index = 0;
 
   // Whether to swap references to always project on the left reference row.
   const bool vertical_mode = intra_mode >= 34;
@@ -418,12 +419,14 @@ static void kvz_intra_pred_planar_generic(
 * \param in_ref_above  Pointer to -1 index of above reference, length=width*2+1.
 * \param in_ref_left   Pointer to -1 index of left reference, length=width*2+1.
 * \param dst           Buffer of size width*width.
+* \param multi_ref_idx Reference line index. May be non-zero when MRL is used.
 */
 static void kvz_intra_pred_filtered_dc_generic(
   const int_fast8_t log2_width,
   const kvz_pixel *const ref_top,
   const kvz_pixel *const ref_left,
-  kvz_pixel *const out_block)
+  kvz_pixel *const out_block,
+  const uint8_t multi_ref_idx)
 {
   assert(log2_width >= 2 && log2_width <= 5);
 
@@ -432,8 +435,8 @@ static void kvz_intra_pred_filtered_dc_generic(
 
   int_fast16_t sum = 0;
   for (int_fast8_t i = 0; i < width; ++i) {
-    sum += ref_top[i + 1];
+    sum += ref_top[i + 1 + multi_ref_idx];
-    sum += ref_left[i + 1];
+    sum += ref_left[i + 1 + multi_ref_idx];
   }
 
   const kvz_pixel dc_val = (sum + width) >> (log2_width + 1);

src/strategies/strategies-intra.h

@@ -49,7 +49,8 @@ typedef void (angular_pred_func)(
   const int_fast8_t channel_type,
   const kvz_pixel *const in_ref_above,
   const kvz_pixel *const in_ref_left,
-  kvz_pixel *const dst);
+  kvz_pixel *const dst,
+  const uint8_t multi_ref_idx);
 
 typedef void (intra_pred_planar_func)(
   const int_fast8_t log2_width,
@@ -61,7 +62,8 @@ typedef void (intra_pred_filtered_dc_func)(
   const int_fast8_t log2_width,
   const kvz_pixel *const ref_top,
   const kvz_pixel *const ref_left,
-  kvz_pixel *const out_block);
+  kvz_pixel *const out_block,
+  const uint8_t multi_ref_idx);
 
 typedef void (pdpc_planar_dc_func)(
   const int mode,