Merge branch 'ibc_fixes'

src/search_ibc.c

@@ -226,7 +226,30 @@ static bool check_mv_cost_satd(ibc_search_info_t *info,
                           double* best_bits,
                           vector2d_t *best_mv)
 {
+  if (!intmv_within_ibc_range(info, x, y)) return false;
 
+  double cost = calculate_ibc_cost_satd(info->state, info->lcu, &info->origin, x, y);
+
+  if (cost >= *best_cost) return false;
+
+  cost += info->mvd_cost_func(
+      info->state,
+      x, y, INTERNAL_MV_PREC,
+      info->mv_cand,
+      NULL,
+      0,
+      0,
+      best_bits
+  );
+
+  if (cost >= *best_cost) return false;
+
+  // Set to motion vector in internal pixel precision.
+  best_mv->x = x * (1 << INTERNAL_MV_PREC);
+  best_mv->y = y * (1 << INTERNAL_MV_PREC);
+  *best_cost = cost;
+
+  return true;
  }
 /**
  * \brief Calculate cost for an integer motion vector.

src/strategies/generic/picture-generic.c

@@ -1374,17 +1374,29 @@ INLINE static uint32_t uvg_crc32c_4_generic(uint32_t crc, const uvg_pixel *buf)
   return crc;
 }
 
+INLINE static uint32_t uvg_crc32c_4_16bit_generic(uint32_t crc, const uvg_pixel *buf)
+{
+  const uint8_t *buf8 = (const uint8_t *)buf;
+  for (int i = 0; i < 8; i++) {
+    crc = (crc >> 8) ^ uvg_crc_table[(crc ^ buf8[i]) & 0xFF];
+  }
+  return crc;
+}
 
 INLINE static uint32_t uvg_crc32c_8_generic(uint32_t crc, const uvg_pixel *buf)
 {
-  crc = (crc >> 8) ^ uvg_crc_table[(crc ^ buf[0]) & 0xFF];
+  for (int i = 0; i < 8; i++) {
-  crc = (crc >> 8) ^ uvg_crc_table[(crc ^ buf[1]) & 0xFF];
+    crc = (crc >> 8) ^ uvg_crc_table[(crc ^ buf[i]) & 0xFF];
-  crc = (crc >> 8) ^ uvg_crc_table[(crc ^ buf[2]) & 0xFF];
+  }
-  crc = (crc >> 8) ^ uvg_crc_table[(crc ^ buf[3]) & 0xFF];
+  return crc;
-  crc = (crc >> 8) ^ uvg_crc_table[(crc ^ buf[4]) & 0xFF];
+}
-  crc = (crc >> 8) ^ uvg_crc_table[(crc ^ buf[5]) & 0xFF];
+
-  crc = (crc >> 8) ^ uvg_crc_table[(crc ^ buf[6]) & 0xFF];
+INLINE static uint32_t uvg_crc32c_8_16bit_generic(uint32_t crc, const uvg_pixel *buf)
-  crc = (crc >> 8) ^ uvg_crc_table[(crc ^ buf[7]) & 0xFF];
+{
+  const uint8_t* buf8 = (const uint8_t*)buf;
+  for (int i = 0; i < 16; i++) {
+    crc = (crc >> 8) ^ uvg_crc_table[(crc ^ buf8[i]) & 0xFF];
+  }
   return crc;
 }
 
@@ -1401,34 +1413,29 @@ static uint32_t uvg_crc32c_4x4_8bit_generic(const uvg_pixel *buf, uint32_t pic_s
 static uint32_t uvg_crc32c_4x4_16bit_generic(const uvg_pixel *buf, uint32_t pic_stride)
 {
   uint32_t crc = 0xFFFFFFFF;
-  crc = uvg_crc32c_4_generic(crc, &buf[0 * pic_stride]);
+  crc = uvg_crc32c_4_16bit_generic(crc, &buf[0 * pic_stride]);
-  crc = uvg_crc32c_4_generic(crc, &buf[0 * pic_stride] + 4);
+  crc = uvg_crc32c_4_16bit_generic(crc, &buf[1 * pic_stride]);
+  crc = uvg_crc32c_4_16bit_generic(crc, &buf[2 * pic_stride]);
+  crc = uvg_crc32c_4_16bit_generic(crc, &buf[3 * pic_stride]);
 
-  crc = uvg_crc32c_4_generic(crc, &buf[1 * pic_stride]);
-  crc = uvg_crc32c_4_generic(crc, &buf[1 * pic_stride] + 4);
-
-  crc = uvg_crc32c_4_generic(crc, &buf[2 * pic_stride]);
-  crc = uvg_crc32c_4_generic(crc, &buf[2 * pic_stride] + 4);
-
-  crc = uvg_crc32c_4_generic(crc, &buf[3 * pic_stride]);
-  crc = uvg_crc32c_4_generic(crc, &buf[3 * pic_stride] + 4);  
   return crc ^ 0xFFFFFFFF;
 }
 
 static uint32_t uvg_crc32c_8x8_8bit_generic(const uvg_pixel *buf, uint32_t pic_stride)
 {
   uint32_t crc = 0xFFFFFFFF;
-  crc = uvg_crc32c_8_generic(crc, &buf[0 * pic_stride]);
+  for (int i = 0; i < 8; i++) {
-  crc = uvg_crc32c_8_generic(crc, &buf[1 * pic_stride]);
+    crc = uvg_crc32c_8_generic(crc, &buf[i * pic_stride]);
+  }
+  return crc ^ 0xFFFFFFFF;
+}
 
-  crc = uvg_crc32c_8_generic(crc, &buf[2 * pic_stride]);
+static uint32_t uvg_crc32c_8x8_16bit_generic(const uvg_pixel *buf, uint32_t pic_stride)
-  crc = uvg_crc32c_8_generic(crc, &buf[3 * pic_stride]);
+{
-
+  uint32_t crc = 0xFFFFFFFF;
-  crc = uvg_crc32c_8_generic(crc, &buf[4 * pic_stride]);
+  for (int i = 0; i < 8; i++) {
-  crc = uvg_crc32c_8_generic(crc, &buf[5 * pic_stride]);
+    crc = uvg_crc32c_8_16bit_generic(crc, &buf[i * pic_stride]);
-
+  }
-  crc = uvg_crc32c_8_generic(crc, &buf[6 * pic_stride]);
-  crc = uvg_crc32c_8_generic(crc, &buf[7 * pic_stride]);  
   return crc ^ 0xFFFFFFFF;
 }
 
@@ -1440,6 +1447,7 @@ int uvg_strategy_register_picture_generic(void* opaque, uint8_t bitdepth)
     success &= uvg_strategyselector_register(opaque, "crc32c_8x8", "generic", 0, &uvg_crc32c_8x8_8bit_generic);
   } else {
     success &= uvg_strategyselector_register(opaque, "crc32c_4x4", "generic", 0, &uvg_crc32c_4x4_16bit_generic);
+    success &= uvg_strategyselector_register(opaque, "crc32c_8x8", "generic", 0, &uvg_crc32c_8x8_16bit_generic);
   }
   
 

src/strategies/sse42/picture-sse42.c

@@ -67,14 +67,20 @@ static uint32_t uvg_crc32c_4x4_16bit_sse42(const uvg_pixel *buf, uint32_t pic_st
 static uint32_t uvg_crc32c_8x8_8bit_sse42(const uvg_pixel *buf, uint32_t pic_stride)
 {
   uint64_t crc = 0xFFFFFFFF;
-  crc = _mm_crc32_u64(crc, *((uint64_t *)&buf[0 * pic_stride]));
+  for (int i = 0; i < 8; i++) {
-  crc = _mm_crc32_u64(crc, *((uint64_t *)&buf[1 * pic_stride]));
+    crc = _mm_crc32_u64(crc, *((uint64_t *)&buf[i * pic_stride]));
-  crc = _mm_crc32_u64(crc, *((uint64_t *)&buf[2 * pic_stride]));
+  }
-  crc = _mm_crc32_u64(crc, *((uint64_t *)&buf[3 * pic_stride]));
+  return (uint32_t)(crc ^ 0xFFFFFFFF);
-  crc = _mm_crc32_u64(crc, *((uint64_t *)&buf[4 * pic_stride]));
+}
-  crc = _mm_crc32_u64(crc, *((uint64_t *)&buf[5 * pic_stride]));
+
-  crc = _mm_crc32_u64(crc, *((uint64_t *)&buf[6 * pic_stride]));
+static uint32_t uvg_crc32c_8x8_16bit_sse42(const uvg_pixel *buf, uint32_t pic_stride)
-  crc = _mm_crc32_u64(crc, *((uint64_t *)&buf[7 * pic_stride]));
+{
+  uint64_t crc = 0xFFFFFFFF;
+  for (int i = 0; i < 8; i++) {
+    crc = _mm_crc32_u64(crc, *((uint64_t *)&buf[i * pic_stride]));
+    crc = _mm_crc32_u64(crc, *((uint64_t *)&buf[i * pic_stride + 4]));
+  }
+
   return (uint32_t)(crc ^ 0xFFFFFFFF);
 }
 
@@ -89,6 +95,7 @@ int uvg_strategy_register_picture_sse42(void* opaque, uint8_t bitdepth) {
     success &= uvg_strategyselector_register(opaque, "crc32c_8x8", "sse42", 0, &uvg_crc32c_8x8_8bit_sse42); 
   } else {
     success &= uvg_strategyselector_register(opaque, "crc32c_4x4", "sse42", 0, &uvg_crc32c_4x4_16bit_sse42); 
+    success &= uvg_strategyselector_register(opaque, "crc32c_8x8", "sse42", 0, &uvg_crc32c_8x8_16bit_sse42); 
   }
 #endif
   return success;

src/strategies/strategies-picture.h

@@ -207,7 +207,6 @@ int uvg_strategy_register_picture(void* opaque, uint8_t bitdepth);
 cost_pixel_nxn_multi_func * uvg_pixels_get_satd_dual_func(unsigned width, unsigned height);
 cost_pixel_nxn_multi_func * uvg_pixels_get_sad_dual_func(unsigned width, unsigned height);
 
-#if UVG_BIT_DEPTH == 8 
 #define STRATEGIES_PICTURE_EXPORTS \
   {"crc32c_4x4", (void**) &uvg_crc32c_4x4}, \
   {"crc32c_8x8", (void **)&uvg_crc32c_8x8}, \
@@ -243,42 +242,5 @@ cost_pixel_nxn_multi_func * uvg_pixels_get_sad_dual_func(unsigned width, unsigne
   {"pixel_var", (void**) &uvg_pixel_var}, \
   {"generate_residual", (void**) &uvg_generate_residual}, \
 
-#else
-#define STRATEGIES_PICTURE_EXPORTS \
-  {"reg_sad", (void**) &uvg_reg_sad}, \
-  {"sad_4x4", (void**) &uvg_sad_4x4}, \
-  {"sad_8x8", (void**) &uvg_sad_8x8}, \
-  {"sad_16x16", (void**) &uvg_sad_16x16}, \
-  {"sad_32x32", (void**) &uvg_sad_32x32}, \
-  {"sad_64x64", (void**) &uvg_sad_64x64}, \
-  {"satd_4x4", (void**) &uvg_satd_4x4}, \
-  {"satd_8x8", (void**) &uvg_satd_8x8}, \
-  {"satd_16x16", (void**) &uvg_satd_16x16}, \
-  {"satd_32x32", (void**) &uvg_satd_32x32}, \
-  {"satd_64x64", (void**) &uvg_satd_64x64}, \
-  {"satd_any_size", (void**) &uvg_satd_any_size}, \
-  {"satd_any_size_vtm", (void**) &uvg_satd_any_size_vtm}, \
-  {"sad_4x4_dual", (void**) &uvg_sad_4x4_dual}, \
-  {"sad_8x8_dual", (void**) &uvg_sad_8x8_dual}, \
-  {"sad_16x16_dual", (void**) &uvg_sad_16x16_dual}, \
-  {"sad_32x32_dual", (void**) &uvg_sad_32x32_dual}, \
-  {"sad_64x64_dual", (void**) &uvg_sad_64x64_dual}, \
-  {"satd_4x4_dual", (void**) &uvg_satd_4x4_dual}, \
-  {"satd_8x8_dual", (void**) &uvg_satd_8x8_dual}, \
-  {"satd_16x16_dual", (void**) &uvg_satd_16x16_dual}, \
-  {"satd_32x32_dual", (void**) &uvg_satd_32x32_dual}, \
-  {"satd_64x64_dual", (void**) &uvg_satd_64x64_dual}, \
-  {"satd_any_size_quad", (void**) &uvg_satd_any_size_quad}, \
-  {"pixels_calc_ssd", (void**) &uvg_pixels_calc_ssd}, \
-  {"bipred_average", (void**) &uvg_bipred_average}, \
-  {"get_optimized_sad", (void**) &uvg_get_optimized_sad}, \
-  {"ver_sad", (void**) &uvg_ver_sad}, \
-  {"hor_sad", (void**) &uvg_hor_sad}, \
-  {"pixel_var", (void**) &uvg_pixel_var}, \
-  {"generate_residual", (void**) &uvg_generate_residual}, \
-
-#endif 
-
-
 
 #endif //STRATEGIES_PICTURE_H_