From: Jingning Han Date: Fri, 1 Nov 2013 19:53:37 +0000 (-0700) Subject: Allocate dual buffer sets for encoding X-Git-Tag: v1.3.0~48^2 X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=3b3aea68344c6bb3107b8189d57c671f4101425f;p=libvpx Allocate dual buffer sets for encoding Allocate memory space of dual buffer sets that store the coeff, qcoeff, dqcoeff, and eobs. Connect the pointers of macroblock_plane and macroblockd_plane to the actual buffer in use accordingly. Change-Id: I2f0b5f482ca879fae39095013eaf8901db20a5a4 --- diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h index 8033a4d15..4d9a92a71 100644 --- a/vp9/encoder/vp9_block.h +++ b/vp9/encoder/vp9_block.h @@ -27,6 +27,17 @@ typedef struct { typedef struct { MODE_INFO mic; uint8_t *zcoeff_blk; + int16_t *coeff[MAX_MB_PLANE][2]; + int16_t *qcoeff[MAX_MB_PLANE][2]; + int16_t *dqcoeff[MAX_MB_PLANE][2]; + uint16_t *eobs[MAX_MB_PLANE][2]; + + // dual buffer pointers + int16_t *coeff_pbuf[MAX_MB_PLANE][2]; + int16_t *qcoeff_pbuf[MAX_MB_PLANE][2]; + int16_t *dqcoeff_pbuf[MAX_MB_PLANE][2]; + uint16_t *eobs_pbuf[MAX_MB_PLANE][2]; + int num_4x4_blk; int skip; int_mv best_ref_mv; @@ -57,7 +68,7 @@ typedef struct { struct macroblock_plane { DECLARE_ALIGNED(16, int16_t, src_diff[64 * 64]); - DECLARE_ALIGNED(16, int16_t, coeff[64 * 64]); + int16_t *coeff; struct buf_2d src; // Quantizer setings diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c index 702fc70bb..094c0287c 100644 --- a/vp9/encoder/vp9_encodeframe.c +++ b/vp9/encoder/vp9_encodeframe.c @@ -367,6 +367,8 @@ static void update_state(VP9_COMP *cpi, PICK_MODE_CONTEXT *ctx, VP9_COMMON *const cm = &cpi->common; MACROBLOCK *const x = &cpi->mb; MACROBLOCKD *const xd = &x->e_mbd; + struct macroblock_plane *const p = x->plane; + struct macroblockd_plane *const pd = xd->plane; MODE_INFO *mi = &ctx->mic; MB_MODE_INFO *const mbmi = &xd->mi_8x8[0]->mbmi; MODE_INFO *mi_addr = xd->mi_8x8[0]; @@ -383,6 +385,13 @@ static void update_state(VP9_COMP *cpi, PICK_MODE_CONTEXT *ctx, *mi_addr = *mi; + for (i = 0; i < MAX_MB_PLANE; ++i) { + p[i].coeff = ctx->coeff_pbuf[i][1]; + pd[i].qcoeff = ctx->qcoeff_pbuf[i][1]; + pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][1]; + pd[i].eobs = ctx->eobs_pbuf[i][1]; + } + // Restore the coding context of the MB to that that was in place // when the mode was picked for it for (y = 0; y < mi_height; y++) @@ -578,6 +587,9 @@ static void pick_sb_modes(VP9_COMP *cpi, const TileInfo *const tile, VP9_COMMON *const cm = &cpi->common; MACROBLOCK *const x = &cpi->mb; MACROBLOCKD *const xd = &x->e_mbd; + struct macroblock_plane *const p = x->plane; + struct macroblockd_plane *const pd = xd->plane; + int i; int orig_rdmult = x->rdmult; double rdmult_ratio; @@ -600,6 +612,13 @@ static void pick_sb_modes(VP9_COMP *cpi, const TileInfo *const tile, set_offsets(cpi, tile, mi_row, mi_col, bsize); xd->mi_8x8[0]->mbmi.sb_type = bsize; + for (i = 0; i < MAX_MB_PLANE; ++i) { + p[i].coeff = ctx->coeff_pbuf[i][1]; + pd[i].qcoeff = ctx->qcoeff_pbuf[i][1]; + pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][1]; + pd[i].eobs = ctx->eobs_pbuf[i][1]; + } + // Set to zero to make sure we do not use the previous encoded frame stats xd->mi_8x8[0]->mbmi.skip_coeff = 0; diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c index 6a3555d68..b5428df28 100644 --- a/vp9/encoder/vp9_firstpass.c +++ b/vp9/encoder/vp9_firstpass.c @@ -482,6 +482,10 @@ void vp9_first_pass(VP9_COMP *cpi) { VP9_COMMON *const cm = &cpi->common; MACROBLOCKD *const xd = &x->e_mbd; TileInfo tile; + struct macroblock_plane *const p = x->plane; + struct macroblockd_plane *const pd = xd->plane; + PICK_MODE_CONTEXT *ctx = &x->sb64_context; + int i; int recon_yoffset, recon_uvoffset; const int lst_yv12_idx = cm->ref_frame_map[cpi->lst_fb_idx]; @@ -525,6 +529,14 @@ void vp9_first_pass(VP9_COMP *cpi) { vp9_frame_init_quantizer(cpi); + for (i = 0; i < MAX_MB_PLANE; ++i) { + p[i].coeff = ctx->coeff_pbuf[i][1]; + pd[i].qcoeff = ctx->qcoeff_pbuf[i][1]; + pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][1]; + pd[i].eobs = ctx->eobs_pbuf[i][1]; + } + + // Initialise the MV cost table to the defaults // if( cm->current_video_frame == 0) // if ( 0 ) diff --git a/vp9/encoder/vp9_onyx_if.c b/vp9/encoder/vp9_onyx_if.c index 4f1357a2d..f4106934c 100644 --- a/vp9/encoder/vp9_onyx_if.c +++ b/vp9/encoder/vp9_onyx_if.c @@ -1436,6 +1436,49 @@ static void cal_nmvsadcosts_hp(int *mvsadcost[2]) { } while (++i <= MV_MAX); } +static void alloc_mode_context(VP9_COMMON *cm, int num_4x4_blk, + PICK_MODE_CONTEXT *ctx) { + int num_pix = num_4x4_blk << 4; + int i, k; + ctx->num_4x4_blk = num_4x4_blk; + CHECK_MEM_ERROR(cm, ctx->zcoeff_blk, + vpx_calloc(num_4x4_blk, sizeof(uint8_t))); + for (i = 0; i < MAX_MB_PLANE; ++i) { + for (k = 0; k < 2; ++k) { + CHECK_MEM_ERROR(cm, ctx->coeff[i][k], + vpx_memalign(16, num_pix * sizeof(int16_t))); + CHECK_MEM_ERROR(cm, ctx->qcoeff[i][k], + vpx_memalign(16, num_pix * sizeof(int16_t))); + CHECK_MEM_ERROR(cm, ctx->dqcoeff[i][k], + vpx_memalign(16, num_pix * sizeof(int16_t))); + CHECK_MEM_ERROR(cm, ctx->eobs[i][k], + vpx_memalign(16, num_pix * sizeof(uint16_t))); + ctx->coeff_pbuf[i][k] = ctx->coeff[i][k]; + ctx->qcoeff_pbuf[i][k] = ctx->qcoeff[i][k]; + ctx->dqcoeff_pbuf[i][k] = ctx->dqcoeff[i][k]; + ctx->eobs_pbuf[i][k] = ctx->eobs[i][k]; + } + } +} + +static void free_mode_context(PICK_MODE_CONTEXT *ctx) { + int i, k; + vpx_free(ctx->zcoeff_blk); + ctx->zcoeff_blk = 0; + for (i = 0; i < MAX_MB_PLANE; ++i) { + for (k = 0; k < 2; ++k) { + vpx_free(ctx->coeff[i][k]); + ctx->coeff[i][k] = 0; + vpx_free(ctx->qcoeff[i][k]); + ctx->qcoeff[i][k] = 0; + vpx_free(ctx->dqcoeff[i][k]); + ctx->dqcoeff[i][k] = 0; + vpx_free(ctx->eobs[i][k]); + ctx->eobs[i][k] = 0; + } + } +} + static void init_pick_mode_context(VP9_COMP *cpi) { int i; MACROBLOCK *x = &cpi->mb; @@ -1451,9 +1494,7 @@ static void init_pick_mode_context(VP9_COMP *cpi) { for (xd->mb_index = 0; xd->mb_index < 4; ++xd->mb_index) { for (xd->b_index = 0; xd->b_index < 16 / num_4x4_blk; ++xd->b_index) { PICK_MODE_CONTEXT *ctx = get_block_context(x, i); - ctx->num_4x4_blk = num_4x4_blk; - CHECK_MEM_ERROR(cm, ctx->zcoeff_blk, - vpx_calloc(num_4x4_blk, sizeof(uint8_t))); + alloc_mode_context(cm, num_4x4_blk, ctx); } } } @@ -1463,22 +1504,19 @@ static void init_pick_mode_context(VP9_COMP *cpi) { ++xd->mb_index) { PICK_MODE_CONTEXT *ctx = get_block_context(x, i); ctx->num_4x4_blk = num_4x4_blk; - CHECK_MEM_ERROR(cm, ctx->zcoeff_blk, - vpx_calloc(num_4x4_blk, sizeof(uint8_t))); + alloc_mode_context(cm, num_4x4_blk, ctx); } } } else if (i < BLOCK_64X64) { for (xd->sb_index = 0; xd->sb_index < 256 / num_4x4_blk; ++xd->sb_index) { PICK_MODE_CONTEXT *ctx = get_block_context(x, i); ctx->num_4x4_blk = num_4x4_blk; - CHECK_MEM_ERROR(cm, ctx->zcoeff_blk, - vpx_calloc(num_4x4_blk, sizeof(uint8_t))); + alloc_mode_context(cm, num_4x4_blk, ctx); } } else { PICK_MODE_CONTEXT *ctx = get_block_context(x, i); ctx->num_4x4_blk = num_4x4_blk; - CHECK_MEM_ERROR(cm, ctx->zcoeff_blk, - vpx_calloc(num_4x4_blk, sizeof(uint8_t))); + alloc_mode_context(cm, num_4x4_blk, ctx); } } } @@ -1496,8 +1534,7 @@ static void free_pick_mode_context(MACROBLOCK *x) { for (xd->mb_index = 0; xd->mb_index < 4; ++xd->mb_index) { for (xd->b_index = 0; xd->b_index < 16 / num_4x4_blk; ++xd->b_index) { PICK_MODE_CONTEXT *ctx = get_block_context(x, i); - vpx_free(ctx->zcoeff_blk); - ctx->zcoeff_blk = 0; + free_mode_context(ctx); } } } @@ -1506,35 +1543,21 @@ static void free_pick_mode_context(MACROBLOCK *x) { for (xd->mb_index = 0; xd->mb_index < 64 / num_4x4_blk; ++xd->mb_index) { PICK_MODE_CONTEXT *ctx = get_block_context(x, i); - vpx_free(ctx->zcoeff_blk); - ctx->zcoeff_blk = 0; + free_mode_context(ctx); } } } else if (i < BLOCK_64X64) { for (xd->sb_index = 0; xd->sb_index < 256 / num_4x4_blk; ++xd->sb_index) { PICK_MODE_CONTEXT *ctx = get_block_context(x, i); - vpx_free(ctx->zcoeff_blk); - ctx->zcoeff_blk = 0; + free_mode_context(ctx); } } else { PICK_MODE_CONTEXT *ctx = get_block_context(x, i); - vpx_free(ctx->zcoeff_blk); - ctx->zcoeff_blk = 0; + free_mode_context(ctx); } } } -static void init_macroblock(VP9_COMP *const cpi) { - MACROBLOCKD *xd = &cpi->mb.e_mbd; - struct macroblockd_plane *const pd = xd->plane; - int i; - for (i = 0; i < MAX_MB_PLANE; ++i) { - pd[i].qcoeff = cpi->qcoeff[i]; - pd[i].dqcoeff = cpi->dqcoeff[i]; - pd[i].eobs = cpi->eobs[i]; - } -} - VP9_PTR vp9_create_compressor(VP9_CONFIG *oxcf) { int i, j; volatile union { @@ -1573,8 +1596,6 @@ VP9_PTR vp9_create_compressor(VP9_CONFIG *oxcf) { init_pick_mode_context(cpi); - init_macroblock(cpi); - cm->current_video_frame = 0; cpi->kf_overspend_bits = 0; cpi->kf_bitrate_adjustment = 0; diff --git a/vp9/encoder/vp9_onyx_int.h b/vp9/encoder/vp9_onyx_int.h index 839a92b7e..9e802123c 100644 --- a/vp9/encoder/vp9_onyx_int.h +++ b/vp9/encoder/vp9_onyx_int.h @@ -312,11 +312,6 @@ typedef struct VP9_COMP { VP9_COMMON common; VP9_CONFIG oxcf; struct rdcost_block_args rdcost_stack; - - DECLARE_ALIGNED(16, int16_t, qcoeff[MAX_MB_PLANE][64 * 64]); - DECLARE_ALIGNED(16, int16_t, dqcoeff[MAX_MB_PLANE][64 * 64]); - DECLARE_ALIGNED(16, uint16_t, eobs[MAX_MB_PLANE][256]); - struct lookahead_ctx *lookahead; struct lookahead_entry *source; #if CONFIG_MULTIPLE_ARF diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c index 6aaa2c77d..2d3bebcc7 100644 --- a/vp9/encoder/vp9_rdopt.c +++ b/vp9/encoder/vp9_rdopt.c @@ -1901,6 +1901,7 @@ static void rd_check_segment_txsize(VP9_COMP *cpi, MACROBLOCK *x, bsi->ref_mv, bsi->second_ref_mv, x->nmvjointcost, x->mvcost, cpi); + bsi->rdstat[i][mode_idx].mvs[0].as_int = mode_mv[this_mode].as_int; if (num_4x4_blocks_wide > 1) bsi->rdstat[i + 1][mode_idx].mvs[0].as_int =