hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-27 19:24:06 +00:00
Code Issues Actions Projects Releases Wiki Activity

Add TR_MAX_WIDTH.

Browse source 
- Max transform size is constrained by but independent of LCU size.

- Luma and chroma now have the same stride for transform arrays.
This commit is contained in:
Ari Koivula 2014-05-13 10:04:05 +03:00
parent 1c38209cab
commit 637aceb495
2 changed files with 16 additions and 13 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
src/global.h
View file

@ -86,7 +86,10 @@ typedef int16_t coefficient;
#define CU_MIN_SIZE_PIXELS (1 << MIN_SIZE) /*!< pow(2, MIN_SIZE) */ #define CU_MIN_SIZE_PIXELS (1 << MIN_SIZE) /*!< pow(2, MIN_SIZE) */
#define LCU_WIDTH (1 << (MIN_SIZE + MAX_DEPTH)) /*!< spec: CtbSizeY */ #define LCU_WIDTH (1 << (MIN_SIZE + MAX_DEPTH)) /*!< spec: CtbSizeY */
#define LCU_WIDTH_C (LCU_WIDTH / 2) /*!< CtbWidthC and CtbHeightC */ #define LCU_WIDTH_C (LCU_WIDTH / 2) /*!< spec: CtbWidthC and CtbHeightC */
#define TR_MAX_LOG2_SIZE 5 /*!< spec: Log2MaxTrafoSize <= Min(CtbLog2SizeY, 5) */
#define TR_MAX_WIDTH (1 << 5) /*!< spec: Log2MaxTrafoSize */
#if LCU_WIDTH != 64 #if LCU_WIDTH != 64
#error "Kvazaar only support LCU_WIDTH == 64" #error "Kvazaar only support LCU_WIDTH == 64"

24
src/transform.c
View file

@ -865,11 +865,11 @@ int quantize_residual_chroma(encoder_state * const encoder_state,
cu_info *cur_cu, int luma_depth, color_index color, cu_info *cur_cu, int luma_depth, color_index color,
const pixel *base_u, pixel *recbase_u, coefficient *orig_coeff_u) const pixel *base_u, pixel *recbase_u, coefficient *orig_coeff_u)
{ {
pixel pred_u[LCU_WIDTH*LCU_WIDTH>>2]; pixel pred_u[TR_MAX_WIDTH * TR_MAX_WIDTH];
coefficient coeff_u[LCU_WIDTH*LCU_WIDTH>>2]; coefficient coeff_u[TR_MAX_WIDTH * TR_MAX_WIDTH];
int16_t block[LCU_WIDTH*LCU_WIDTH>>2]; int16_t block[TR_MAX_WIDTH * TR_MAX_WIDTH];
int16_t pre_quant_coeff[LCU_WIDTH*LCU_WIDTH>>2]; int16_t pre_quant_coeff[TR_MAX_WIDTH * TR_MAX_WIDTH];
const int chroma_depth = (luma_depth == MAX_PU_DEPTH ? luma_depth - 1 : luma_depth); const int chroma_depth = (luma_depth == MAX_PU_DEPTH ? luma_depth - 1 : luma_depth);
const int8_t width_c = LCU_WIDTH >> (chroma_depth + 1); const int8_t width_c = LCU_WIDTH >> (chroma_depth + 1);
@ -1056,13 +1056,13 @@ void encode_transform_tree(encoder_state * const encoder_state, int32_t x, int32
coefficient *orig_coeff_y = &lcu->coeff.y[luma_offset]; coefficient *orig_coeff_y = &lcu->coeff.y[luma_offset];
// Temporary buffers. Not really used for much. Possibly unnecessary. // Temporary buffers. Not really used for much. Possibly unnecessary.
pixel pred_y[LCU_WIDTH*LCU_WIDTH]; pixel pred_y[TR_MAX_WIDTH * TR_MAX_WIDTH];
// Buffers for coefficients. // Buffers for coefficients.
coefficient coeff_y[LCU_WIDTH*LCU_WIDTH]; coefficient coeff_y[TR_MAX_WIDTH * TR_MAX_WIDTH];
// Temporary buffers for kvantization and transformation. // Temporary buffers for kvantization and transformation.
int16_t block[LCU_WIDTH*LCU_WIDTH>>2]; int16_t block[TR_MAX_WIDTH * TR_MAX_WIDTH];
int16_t pre_quant_coeff[LCU_WIDTH*LCU_WIDTH>>2]; int16_t pre_quant_coeff[TR_MAX_WIDTH * TR_MAX_WIDTH];
uint32_t ac_sum = 0; uint32_t ac_sum = 0;
uint8_t scan_idx_luma = SCAN_DIAG; uint8_t scan_idx_luma = SCAN_DIAG;
@ -1092,7 +1092,7 @@ void encode_transform_tree(encoder_state * const encoder_state, int32_t x, int32
// Copy Luma and Chroma to the pred-block // Copy Luma and Chroma to the pred-block
for(y = 0; y < width; y++) { for(y = 0; y < width; y++) {
for(x = 0; x < width; x++) { for(x = 0; x < width; x++) {
pred_y[x+y*LCU_WIDTH]=recbase_y[x+y*LCU_WIDTH]; pred_y[x+y*TR_MAX_WIDTH]=recbase_y[x+y*LCU_WIDTH];
} }
} }
@ -1103,7 +1103,7 @@ void encode_transform_tree(encoder_state * const encoder_state, int32_t x, int32
for (y = 0; y < width; y++) { for (y = 0; y < width; y++) {
for (x = 0; x < width; x++) { for (x = 0; x < width; x++) {
block[i] = ((int16_t)base_y[x + y * LCU_WIDTH]) - block[i] = ((int16_t)base_y[x + y * LCU_WIDTH]) -
pred_y[x + y * LCU_WIDTH]; pred_y[x + y * TR_MAX_WIDTH];
#if OPTIMIZATION_SKIP_RESIDUAL_ON_THRESHOLD #if OPTIMIZATION_SKIP_RESIDUAL_ON_THRESHOLD
residual_sum += block[i]; residual_sum += block[i];
#endif #endif
@ -1174,7 +1174,7 @@ void encode_transform_tree(encoder_state * const encoder_state, int32_t x, int32
for (y = 0; y < width; y++) { for (y = 0; y < width; y++) {
for (x = 0; x < width; x++) { for (x = 0; x < width; x++) {
int val = block[i++] + pred_y[x + y * LCU_WIDTH]; int val = block[i++] + pred_y[x + y * TR_MAX_WIDTH];
//TODO: support 10+bits //TODO: support 10+bits
recbase_y[x + y * LCU_WIDTH] = (pixel)CLIP(0, 255, val); recbase_y[x + y * LCU_WIDTH] = (pixel)CLIP(0, 255, val);
} }
@ -1185,7 +1185,7 @@ void encode_transform_tree(encoder_state * const encoder_state, int32_t x, int32
// If luma is 4x4, do chroma for the 8x8 luma area when handling the top // If luma is 4x4, do chroma for the 8x8 luma area when handling the top
// left PU because the coordinates are correct. // left PU because the coordinates are correct.
if (depth <= MAX_DEPTH || pu_index == 0) { if (depth <= MAX_DEPTH || pu_index == 0) {
const int chroma_offset = lcu_px.x / 2 + lcu_px.y / 2 * LCU_WIDTH / 2; const int chroma_offset = lcu_px.x / 2 + lcu_px.y / 2 * LCU_WIDTH_C;
pixel *recbase_u = &lcu->rec.u[chroma_offset]; pixel *recbase_u = &lcu->rec.u[chroma_offset];
pixel *recbase_v = &lcu->rec.v[chroma_offset]; pixel *recbase_v = &lcu->rec.v[chroma_offset];
const pixel *base_u = &lcu->ref.u[chroma_offset]; const pixel *base_u = &lcu->ref.u[chroma_offset];