From 5264569f43b9a4bb58ad0661c6ed1b3447126175 Mon Sep 17 00:00:00 2001 From: Marko Viitanen Date: Thu, 19 Sep 2013 15:08:30 +0300 Subject: [PATCH] Refactor: inter.c/.h full cleanup --- src/inter.c | 263 +++++++++++++++++++++++++++------------------------ src/inter.h | 2 +- src/search.c | 6 +- 3 files changed, 142 insertions(+), 129 deletions(-) diff --git a/src/inter.c b/src/inter.c index a8e88a5e..c82999eb 100644 --- a/src/inter.c +++ b/src/inter.c @@ -18,153 +18,164 @@ #include "config.h" -/*! - \brief Set block mode (and init typedata) - \param pic picture to use - \param xCtb x CU position (smallest CU) - \param yCtb y CU position (smallest CU) - \param depth current CU depth - \param mode mode to set - \returns Void +/** + * \brief Set block info to the CU structure + * \param pic picture to use + * \param x_cu x CU position (smallest CU) + * \param y_cu y CU position (smallest CU) + * \param depth current CU depth + * \param cur_cu CU to take the settings from + * \returns Void */ -void inter_setBlockMode(picture* pic,uint32_t x_cu, uint32_t y_cu, uint8_t depth, CU_info* cur_cu) +void inter_set_block(picture* pic, uint32_t x_cu, uint32_t y_cu, uint8_t depth, CU_info* cur_cu) { uint32_t x,y,d; - /* Width in smallest CU */ - int width_in_SCU = pic->width_in_lcu<>depth)/(LCU_WIDTH>>MAX_DEPTH); - for(y = y_cu; y < y_cu+block_SCU_width; y++) - { - int CUpos = y*width_in_SCU; - for(x = x_cu; x < x_cu+block_SCU_width; x++) - { - for(d = 0; d < MAX_DEPTH+1; d++) - { - pic->CU[d][CUpos+x].depth = depth; - pic->CU[d][CUpos+x].type = CU_INTER; - pic->CU[d][CUpos+x].inter.mode = cur_cu->inter.mode; - pic->CU[d][CUpos+x].inter.mv[0] = cur_cu->inter.mv[0]; - pic->CU[d][CUpos+x].inter.mv[1] = cur_cu->inter.mv[1]; - pic->CU[d][CUpos+x].inter.mv_dir = cur_cu->inter.mv_dir; + // Width in smallest CU + int width_in_scu = pic->width_in_lcu<>depth)/(LCU_WIDTH>>MAX_DEPTH); + // Loop through all the block in the area of cur_cu + for (y = y_cu; y < y_cu + block_scu_width; y++) { + int cu_pos = y * width_in_scu; //!< calculate y-position once, use with every x + for (x = x_cu; x < x_cu + block_scu_width; x++) { + // reset all depths to the same MV/inter data + for(d = 0; d < MAX_DEPTH + 1; d++) { + pic->CU[d][cu_pos + x].depth = depth; + pic->CU[d][cu_pos + x].type = CU_INTER; + pic->CU[d][cu_pos + x].inter.mode = cur_cu->inter.mode; + pic->CU[d][cu_pos + x].inter.mv[0] = cur_cu->inter.mv[0]; + pic->CU[d][cu_pos + x].inter.mv[1] = cur_cu->inter.mv[1]; + pic->CU[d][cu_pos + x].inter.mv_dir = cur_cu->inter.mv_dir; } } } } -/*! - \brief Reconstruct inter block - \param ref picture to copy the data from - \param xpos block x position - \param ypos block y position - \param width block width - \param mv[2] motion vector - \param dst destination picture - \returns Void +/** + * \brief Reconstruct inter block + * \param ref picture to copy the data from + * \param xpos block x position + * \param ypos block y position + * \param width block width + * \param mv[2] motion vector + * \param dst destination picture + * \returns Void */ -void inter_recon(picture* ref,int32_t xpos, int32_t ypos,int32_t width, int16_t mv[2], picture* dst) +void inter_recon(picture* ref,int32_t xpos, int32_t ypos,int32_t width, int16_t mv[2], picture *dst) { int x,y,coord_x,coord_y; - /* negative overflow present */ - int8_t overflow_neg_x = (xpos+mv[0] < 0)?1:0; - int8_t overflow_neg_y = (ypos+mv[1] < 0)?1:0; + int32_t dst_width_c = dst->width>>1; //!< Destination picture width in chroma pixels + int32_t ref_width_c = ref->width>>1; //!< Reference picture width in chroma pixels - /* positive overflow present */ - int8_t overflow_pos_x = (xpos+mv[0]+width > ref->width )?1:0; - int8_t overflow_pos_y = (ypos+mv[1]+width > ref->height)?1:0; + // negative overflow flag + int8_t overflow_neg_x = (xpos + mv[0] < 0)?1:0; + int8_t overflow_neg_y = (ypos + mv[1] < 0)?1:0; - /* TODO: Fractional pixel support */ + // positive overflow flag + int8_t overflow_pos_x = (xpos + mv[0] + width > ref->width )?1:0; + int8_t overflow_pos_y = (ypos + mv[1] + width > ref->height)?1:0; + + // TODO: Fractional pixel support mv[0] = mv[0]>>2; mv[1] = mv[1]>>2; - /* With overflow present, more checking */ + // With overflow present, more checking if (overflow_neg_x || overflow_neg_y || overflow_pos_x || overflow_pos_y) { - /* Copy Luma with boundary checking */ - for (y = ypos; y < ypos+width; y++) { - for (x = xpos; x < xpos+width; x++) { - coord_x = x; - coord_y = y; - overflow_neg_x = (x+mv[0] < 0)?1:0; - overflow_neg_y = (y+mv[1] < 0)?1:0; + // Copy Luma with boundary checking + for (y = ypos; y < ypos + width; y++) { + for (x = xpos; x < xpos + width; x++) { + coord_x = x + mv[0]; + coord_y = y + mv[1]; + overflow_neg_x = (coord_x < 0)?1:0; + overflow_neg_y = (coord_y < 0)?1:0; - overflow_pos_x = (x+mv[0] >= ref->width )?1:0; - overflow_pos_y = (y+mv[1] >= ref->height)?1:0; + overflow_pos_x = (coord_x >= ref->width )?1:0; + overflow_pos_y = (coord_y >= ref->height)?1:0; + // On x-overflow set coord_x accordingly + if (overflow_neg_x) { + coord_x = 0; + } else if (overflow_pos_x) { + coord_x = ref->width - 1; + } + + // On y-overflow set coord_y accordingly + if (overflow_neg_y) { + coord_y = 0; + } else if (overflow_pos_y) { + coord_y = ref->height - 1; + } + + // set destination to (corrected) pixel value from the reference + dst->y_recdata[y * dst->width + x] = ref->y_recdata[coord_y*ref->width + coord_x]; + } + } + + // Copy Chroma with boundary checking + // TODO: chroma fractional pixel interpolation + for (y = ypos>>1; y < (ypos + width)>>1; y++) { + for (x = xpos>>1; x < (xpos + width)>>1; x++) { + coord_x = x + (mv[0]>>1); + coord_y = y + (mv[1]>>1); + overflow_neg_x = (coord_x < 0)?1:0; + overflow_neg_y = (y + (mv[1]>>1) < 0)?1:0; + + overflow_pos_x = (coord_x >= ref->width>>1 )?1:0; + overflow_pos_y = (coord_y >= ref->height>>1)?1:0; + + // On x-overflow set coord_x accordingly if(overflow_neg_x) { - coord_x = -mv[0]; + coord_x = 0; } else if(overflow_pos_x) { - coord_x = ref->width-1-mv[0]; + coord_x = (ref->width>>1) - 1; } + // On y-overflow set coord_y accordingly if(overflow_neg_y) { - coord_y = -mv[1]; + coord_y = 0; } else if(overflow_pos_y) { - coord_y = ref->height-1-mv[1]; + coord_y = (ref->height>>1) - 1; } - dst->y_recdata[y*dst->width+x] = ref->y_recdata[(coord_y+mv[1])*ref->width+(coord_x+mv[0])]; + // set destinations to (corrected) pixel value from the reference + dst->u_recdata[y*dst_width_c + x] = ref->u_recdata[coord_y*ref_width_c + coord_x]; + dst->v_recdata[y*dst_width_c + x] = ref->v_recdata[coord_y*ref_width_c + coord_x]; + } + } + } else { //If no overflow, we can copy without checking boundaries + // Copy Luma + for (y = ypos; y < ypos + width; y++) { + coord_y = (y + mv[1]) * ref->width; // pre-calculate + for (x = xpos; x < xpos + width; x++) { + dst->y_recdata[y * dst->width + x] = ref->y_recdata[coord_y + x + mv[0]]; } } - /* Copy Chroma with boundary checking */ - for (y = ypos>>1; y < (ypos+width)>>1; y++) { - for (x = xpos>>1; x < (xpos+width)>>1; x++) { - coord_x = x; - coord_y = y; - overflow_neg_x = (x+(mv[0]>>1) < 0)?1:0; - overflow_neg_y = (y+(mv[1]>>1) < 0)?1:0; - - overflow_pos_x = (x+(mv[0]>>1) >= ref->width>>1 )?1:0; - overflow_pos_y = (y+(mv[1]>>1) >= ref->height>>1)?1:0; - - if(overflow_neg_x) { - coord_x = -(mv[0]>>1); - } else if(overflow_pos_x) { - coord_x = ((ref->width-mv[0])>>1)-1; - } - - if(overflow_neg_y) { - coord_y = -(mv[1]>>1); - } else if(overflow_pos_y) { - coord_y = ((ref->height-mv[1])>>1)-1; - } - - dst->u_recdata[y*(dst->width>>1)+x] = ref->u_recdata[(coord_y+(mv[1]>>1))*(ref->width>>1)+(coord_x+(mv[0]>>1))]; - dst->v_recdata[y*(dst->width>>1)+x] = ref->v_recdata[(coord_y+(mv[1]>>1))*(ref->width>>1)+(coord_x+(mv[0]>>1))]; - } - } - } else { - /* Copy Luma */ - for (y = ypos; y < ypos+width; y++) { - for (x = xpos; x < xpos+width; x++) { - dst->y_recdata[y*dst->width+x] = ref->y_recdata[(y+mv[1])*ref->width+x+mv[0]]; - } - } - - /* Copy Chroma */ - for (y = ypos>>1; y < (ypos+width)>>1; y++) { - for (x = xpos>>1; x < (xpos+width)>>1; x++) { - dst->u_recdata[y*(dst->width>>1)+x] = ref->u_recdata[(y+(mv[1]>>1))*(ref->width>>1)+x+(mv[0]>>1)]; - dst->v_recdata[y*(dst->width>>1)+x] = ref->v_recdata[(y+(mv[1]>>1))*(ref->width>>1)+x+(mv[0]>>1)]; + // Copy Chroma + // TODO: chroma fractional pixel interpolation + for (y = ypos>>1; y < (ypos + width)>>1; y++) { + coord_y = (y + (mv[1]>>1)) * ref_width_c; // pre-calculate + for (x = xpos>>1; x < (xpos + width)>>1; x++) { + dst->u_recdata[y*dst_width_c + x] = ref->u_recdata[coord_y + x + (mv[0]>>1)]; + dst->v_recdata[y*dst_width_c + x] = ref->v_recdata[coord_y + x + (mv[0]>>1)]; } } } } -/*! - \brief Get MV prediction for current block - \param encoder encoder control struct to use - \param xCtb block x position in SCU - \param yCtb block y position in SCU - \param depth current block depth - \param mv_pred[2][2] 2x motion vector prediction - \returns Void -*/ -void inter_get_mv_cand(encoder_control *encoder,int32_t xCtb, int32_t yCtb,int8_t depth, int16_t mv_cand[2][2]) +/** + * \brief Get MV prediction for current block + * \param encoder encoder control struct to use + * \param x_cu block x position in SCU + * \param y_cu block y position in SCU + * \param depth current block depth + * \param mv_pred[2][2] 2x motion vector prediction + */ +void inter_get_mv_cand(encoder_control *encoder, int32_t x_cu, int32_t y_cu, int8_t depth, int16_t mv_cand[2][2]) { - uint8_t cur_block_in_scu = (LCU_WIDTH>>depth) / CU_MIN_SIZE_PIXELS; + uint8_t cur_block_in_scu = (LCU_WIDTH>>depth) / CU_MIN_SIZE_PIXELS; //!< the width of the current block on SCU uint8_t candidates = 0; - + /* Predictor block locations ____ _______ @@ -179,29 +190,31 @@ void inter_get_mv_cand(encoder_control *encoder,int32_t xCtb, int32_t yCtb,int8_ b0 = b1 = b2 = a0 = a1 = NULL; - if (xCtb != 0) { - a1 = &encoder->in.cur_pic->CU[depth][xCtb-1+(yCtb+cur_block_in_scu-1)*(encoder->in.width_in_lcu<coded) a1 = NULL; + // A0 and A1 availability testing + if (x_cu != 0) { + a1 = &encoder->in.cur_pic->CU[depth][x_cu - 1 + (y_cu + cur_block_in_scu - 1) * (encoder->in.width_in_lcu<coded) a1 = NULL; - if (yCtb+cur_block_in_scu < encoder->in.height_in_lcu<in.cur_pic->CU[depth][xCtb-1+(yCtb+cur_block_in_scu)*(encoder->in.width_in_lcu<coded) a0 = NULL; + if (y_cu + cur_block_in_scu < encoder->in.height_in_lcu<in.cur_pic->CU[depth][x_cu - 1 + (y_cu + cur_block_in_scu) * (encoder->in.width_in_lcu<coded) a0 = NULL; } } - if (yCtb != 0) { - b0 = &encoder->in.cur_pic->CU[depth][xCtb+cur_block_in_scu+(yCtb-1)*(encoder->in.width_in_lcu<in.cur_pic->CU[depth][x_cu + cur_block_in_scu + (y_cu - 1) * (encoder->in.width_in_lcu<coded) b0 = NULL; - b1 = &encoder->in.cur_pic->CU[depth][xCtb+cur_block_in_scu-1+(yCtb-1)*(encoder->in.width_in_lcu<in.cur_pic->CU[depth][x_cu + cur_block_in_scu - 1 + (y_cu - 1) * (encoder->in.width_in_lcu<coded) b1 = NULL; - if (xCtb != 0) { - b2 = &encoder->in.cur_pic->CU[depth][xCtb-1+(yCtb-1)*(encoder->in.width_in_lcu<in.cur_pic->CU[depth][x_cu - 1 + (y_cu - 1) * (encoder->in.width_in_lcu<coded) b2 = NULL; } } - /* Left predictors */ + // Left predictors if (a0 && a0->type == CU_INTER) { mv_cand[candidates][0] = a0->inter.mv[0]; mv_cand[candidates][1] = a0->inter.mv[1]; @@ -212,7 +225,7 @@ void inter_get_mv_cand(encoder_control *encoder,int32_t xCtb, int32_t yCtb,int8_ candidates++; } - /* Top predictors */ + // Top predictors if (b0 && b0->type == CU_INTER) { mv_cand[candidates][0] = b0->inter.mv[0]; mv_cand[candidates][1] = b0->inter.mv[1]; @@ -227,7 +240,7 @@ void inter_get_mv_cand(encoder_control *encoder,int32_t xCtb, int32_t yCtb,int8_ candidates++; } - /* Remove identical candidate */ + // Remove identical candidate if(candidates == 2 && mv_cand[0][0] == mv_cand[1][0] && mv_cand[0][1] == mv_cand[1][1]) { candidates = 1; } @@ -238,10 +251,10 @@ void inter_get_mv_cand(encoder_control *encoder,int32_t xCtb, int32_t yCtb,int8_ } #endif + // Fill with (0,0) while (candidates < 2) { mv_cand[candidates][0] = 0; mv_cand[candidates][1] = 0; candidates++; } - -} \ No newline at end of file +} diff --git a/src/inter.h b/src/inter.h index d5dc6108..de49a541 100644 --- a/src/inter.h +++ b/src/inter.h @@ -18,7 +18,7 @@ #include "encoder.h" -void inter_setBlockMode(picture* pic,uint32_t x_cu, uint32_t y_cu, uint8_t depth, CU_info* cur_cu); +void inter_set_block(picture* pic,uint32_t x_cu, uint32_t y_cu, uint8_t depth, CU_info *cur_cu); void inter_recon(picture *ref,int32_t xpos, int32_t ypos,int32_t width, int16_t mv[2], picture* dst); void inter_get_mv_cand(encoder_control *encoder, int32_t x_cu, int32_t y_cu, int8_t depth, int16_t mv_cand[2][2]); diff --git a/src/search.c b/src/search.c index 028f3635..0209b727 100644 --- a/src/search.c +++ b/src/search.c @@ -215,7 +215,7 @@ void search_tree(encoder_control* encoder,uint16_t xCtb,uint16_t yCtb, uint8_t d cur_CU->type = CU_INTER; cur_CU->inter.mv_dir = 1; - inter_setBlockMode(encoder->in.cur_pic,xCtb,yCtb,depth,cur_CU); + inter_set_block(encoder->in.cur_pic,xCtb,yCtb,depth,cur_CU); } /* INTRA SEARCH */ @@ -284,7 +284,7 @@ uint32_t search_best_mode(encoder_control* encoder,uint16_t xCtb,uint16_t yCtb, { /* Set split to 0 and mode to inter.mode */ picture_setBlockSplit(encoder->in.cur_pic,xCtb,yCtb,depth,0); - inter_setBlockMode(encoder->in.cur_pic,xCtb,yCtb,depth,cur_CU); + inter_set_block(encoder->in.cur_pic,xCtb,yCtb,depth,cur_CU); bestCost = bestInterCost; } /* Else, dont split and recursively set block mode */ @@ -300,7 +300,7 @@ uint32_t search_best_mode(encoder_control* encoder,uint16_t xCtb,uint16_t yCtb, { /* Set split to 0 and mode to inter.mode */ picture_setBlockSplit(encoder->in.cur_pic,xCtb,yCtb,depth,0); - inter_setBlockMode(encoder->in.cur_pic,xCtb,yCtb,depth,cur_CU); + inter_set_block(encoder->in.cur_pic,xCtb,yCtb,depth,cur_CU); bestCost = bestInterCost; } else