--- decoder.c 2002/03/20 00:27:29 1.3 +++ decoder.c 2002/04/04 13:58:06 1.9 @@ -32,6 +32,11 @@ * * History: * + * 29.03.2002 interlacing fix - compensated block wasn't being used when + * reconstructing blocks, thus artifacts + * interlacing speedup - used transfers to re-interlace + * interlaced decoding should be as fast as progressive now + * 26.03.2002 interlacing support - moved transfers outside decode loop * 26.12.2001 decoder_mbinter: dequant/idct moved within if(coded) block * 22.12.2001 block based interpolation * 01.12.2001 inital version; (c)2001 peter ross @@ -68,7 +73,7 @@ { DECODER * dec; - dec = xvid_malloc(sizeof(DECODER), 16); + dec = xvid_malloc(sizeof(DECODER), CACHE_LINE); if (dec == NULL) { return XVID_ERR_MEMORY; @@ -97,7 +102,7 @@ return XVID_ERR_MEMORY; } - dec->mbs = xvid_malloc(sizeof(MACROBLOCK) * dec->mb_width * dec->mb_height, 16); + dec->mbs = xvid_malloc(sizeof(MACROBLOCK) * dec->mb_width * dec->mb_height, CACHE_LINE); if (dec->mbs == NULL) { image_destroy(&dec->cur, dec->edged_width, dec->edged_height); @@ -106,7 +111,6 @@ } init_timer(); - create_vlc_tables(); return XVID_ERR_OK; } @@ -119,8 +123,6 @@ image_destroy(&dec->cur, dec->edged_width, dec->edged_height); xvid_free(dec); - destroy_vlc_tables(); - write_timer(); return XVID_ERR_OK; } @@ -135,36 +137,53 @@ // decode an intra macroblock -void decoder_mbintra(DECODER * dec, MACROBLOCK * mb, int x, int y, uint32_t acpred_flag, uint32_t cbp, Bitstream * bs, int quant, int intra_dc_threshold) +void decoder_mbintra(DECODER * dec, + MACROBLOCK * pMB, + const uint32_t x_pos, + const uint32_t y_pos, + const uint32_t acpred_flag, + const uint32_t cbp, + Bitstream * bs, + const uint32_t quant, + const uint32_t intra_dc_threshold) { - uint32_t k; - for (k = 0; k < 6; k++) + DECLARE_ALIGNED_MATRIX(block, 6, 64, int16_t, CACHE_LINE); + DECLARE_ALIGNED_MATRIX(data, 6, 64, int16_t, CACHE_LINE); + + uint32_t stride = dec->edged_width; + uint32_t stride2 = stride / 2; + uint32_t next_block = stride * 8; + uint32_t i; + uint32_t iQuant = pMB->quant; + uint8_t *pY_Cur, *pU_Cur, *pV_Cur; + + pY_Cur = dec->cur.y + (y_pos << 4) * stride + (x_pos << 4); + pU_Cur = dec->cur.u + (y_pos << 3) * stride2 + (x_pos << 3); + pV_Cur = dec->cur.v + (y_pos << 3) * stride2 + (x_pos << 3); + + memset(block, 0, 6*64*sizeof(int16_t)); // clear + + for (i = 0; i < 6; i++) { - uint32_t dcscalar; - int16_t block[64]; - int16_t data[64]; + uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); int16_t predictors[8]; int start_coeff; - dcscalar = get_dc_scaler(mb->quant, k < 4); - start_timer(); - predict_acdc(dec->mbs, x, y, dec->mb_width, k, block, mb->quant, dcscalar, predictors); + predict_acdc(dec->mbs, x_pos, y_pos, dec->mb_width, i, &block[i*64], iQuant, iDcScaler, predictors); if (!acpred_flag) { - mb->acpred_directions[k] = 0; + pMB->acpred_directions[i] = 0; } stop_prediction_timer(); - memset(block, 0, 64*sizeof(int16_t)); // clear - if (quant < intra_dc_threshold) { int dc_size; int dc_dif; - dc_size = k < 4 ? get_dc_size_lum(bs) : get_dc_size_chrom(bs); + dc_size = i < 4 ? get_dc_size_lum(bs) : get_dc_size_chrom(bs); dc_dif = dc_size ? get_dc_dif(bs, dc_size) : 0 ; if (dc_size > 8) @@ -172,7 +191,7 @@ BitstreamSkip(bs, 1); // marker } - block[0] = dc_dif; + block[i*64 + 0] = dc_dif; start_coeff = 1; } else @@ -181,46 +200,46 @@ } start_timer(); - if (cbp & (1 << (5-k))) // coded + if (cbp & (1 << (5-i))) // coded { - get_intra_block(bs, block, mb->acpred_directions[k], start_coeff); + get_intra_block(bs, &block[i*64], pMB->acpred_directions[i], start_coeff); } stop_coding_timer(); start_timer(); - add_acdc(mb, k, block, dcscalar, predictors); + add_acdc(pMB, i, &block[i*64], iDcScaler, predictors); stop_prediction_timer(); start_timer(); if (dec->quant_type == 0) { - dequant_intra(data, block, mb->quant, dcscalar); + dequant_intra(&data[i*64], &block[i*64], iQuant, iDcScaler); } else { - dequant4_intra(data, block, mb->quant, dcscalar); + dequant4_intra(&data[i*64], &block[i*64], iQuant, iDcScaler); } stop_iquant_timer(); start_timer(); - idct(data); + idct(&data[i*64]); stop_idct_timer(); + } - start_timer(); - if (k < 4) - { - transfer_16to8copy(dec->cur.y + (16*y*dec->edged_width) + 16*x + (4*(k&2)*dec->edged_width) + 8*(k&1), data, dec->edged_width); - } - else if (k == 4) - { - transfer_16to8copy(dec->cur.u+ 8*y*(dec->edged_width/2) + 8*x, data, (dec->edged_width/2)); - } - else // if (k == 5) - { - transfer_16to8copy(dec->cur.v + 8*y*(dec->edged_width/2) + 8*x, data, (dec->edged_width/2)); - } - stop_transfer_timer(); + if (pMB->field_dct) + { + next_block = stride; + stride *= 2; } + + start_timer(); + transfer_16to8copy(pY_Cur, &data[0*64], stride); + transfer_16to8copy(pY_Cur + 8, &data[1*64], stride); + transfer_16to8copy(pY_Cur + next_block, &data[2*64], stride); + transfer_16to8copy(pY_Cur + 8 + next_block, &data[3*64], stride); + transfer_16to8copy(pU_Cur, &data[4*64], stride2); + transfer_16to8copy(pV_Cur, &data[5*64], stride2); + stop_transfer_timer(); } @@ -230,22 +249,41 @@ #define SIGN(X) (((X)>0)?1:-1) #define ABS(X) (((X)>0)?(X):-(X)) static const uint32_t roundtab[16] = - { 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; +{ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; // decode an inter macroblock -void decoder_mbinter(DECODER * dec, MACROBLOCK * mb, int x, int y, uint32_t acpred_flag, uint32_t cbp, Bitstream * bs, int quant, int rounding) +void decoder_mbinter(DECODER * dec, + const MACROBLOCK * pMB, + const uint32_t x_pos, + const uint32_t y_pos, + const uint32_t acpred_flag, + const uint32_t cbp, + Bitstream * bs, + const uint32_t quant, + const uint32_t rounding) { - const uint32_t stride = dec->edged_width; - const uint32_t stride2 = dec->edged_width / 2; + + DECLARE_ALIGNED_MATRIX(block,6, 64, int16_t, CACHE_LINE); + DECLARE_ALIGNED_MATRIX(data, 6, 64, int16_t, CACHE_LINE); + + uint32_t stride = dec->edged_width; + uint32_t stride2 = stride / 2; + uint32_t next_block = stride * 8; + uint32_t i; + uint32_t iQuant = pMB->quant; + uint8_t *pY_Cur, *pU_Cur, *pV_Cur; int uv_dx, uv_dy; - uint32_t k; - if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) + pY_Cur = dec->cur.y + (y_pos << 4) * stride + (x_pos << 4); + pU_Cur = dec->cur.u + (y_pos << 3) * stride2 + (x_pos << 3); + pV_Cur = dec->cur.v + (y_pos << 3) * stride2 + (x_pos << 3); + + if (pMB->mode == MODE_INTER || pMB->mode == MODE_INTER_Q) { - uv_dx = mb->mvs[0].x; - uv_dy = mb->mvs[0].y; + uv_dx = pMB->mvs[0].x; + uv_dy = pMB->mvs[0].y; uv_dx = (uv_dx & 3) ? (uv_dx >> 1) | 1 : uv_dx / 2; uv_dy = (uv_dy & 3) ? (uv_dy >> 1) | 1 : uv_dy / 2; @@ -253,73 +291,75 @@ else { int sum; - sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x; + sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; uv_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) ); - sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y; + sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; uv_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) ); } start_timer(); - interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x, 16*y , mb->mvs[0].x, mb->mvs[0].y, stride, rounding); - interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x + 8, 16*y , mb->mvs[1].x, mb->mvs[1].y, stride, rounding); - interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x, 16*y + 8, mb->mvs[2].x, mb->mvs[2].y, stride, rounding); - interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x + 8, 16*y + 8, mb->mvs[3].x, mb->mvs[3].y, stride, rounding); - interpolate8x8_switch(dec->cur.u, dec->refn.u, 8*x, 8*y, uv_dx, uv_dy, stride2, rounding); - interpolate8x8_switch(dec->cur.v, dec->refn.v, 8*x, 8*y, uv_dx, uv_dy, stride2, rounding); + interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x_pos, 16*y_pos , pMB->mvs[0].x, pMB->mvs[0].y, stride, rounding); + interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x_pos + 8, 16*y_pos , pMB->mvs[1].x, pMB->mvs[1].y, stride, rounding); + interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x_pos, 16*y_pos + 8, pMB->mvs[2].x, pMB->mvs[2].y, stride, rounding); + interpolate8x8_switch(dec->cur.y, dec->refn.y, 16*x_pos + 8, 16*y_pos + 8, pMB->mvs[3].x, pMB->mvs[3].y, stride, rounding); + interpolate8x8_switch(dec->cur.u, dec->refn.u, 8*x_pos, 8*y_pos, uv_dx, uv_dy, stride2, rounding); + interpolate8x8_switch(dec->cur.v, dec->refn.v, 8*x_pos, 8*y_pos, uv_dx, uv_dy, stride2, rounding); stop_comp_timer(); - - for (k = 0; k < 6; k++) + for (i = 0; i < 6; i++) { - int16_t block[64]; - int16_t data[64]; - - if (cbp & (1 << (5-k))) // coded + if (cbp & (1 << (5-i))) // coded { - memset(block, 0, 64 * sizeof(int16_t)); // clear + memset(&block[i*64], 0, 64 * sizeof(int16_t)); // clear start_timer(); - get_inter_block(bs, block); + get_inter_block(bs, &block[i*64]); stop_coding_timer(); start_timer(); if (dec->quant_type == 0) { - dequant_inter(data, block, mb->quant); + dequant_inter(&data[i*64], &block[i*64], iQuant); } else { - dequant4_inter(data, block, mb->quant); + dequant4_inter(&data[i*64], &block[i*64], iQuant); } stop_iquant_timer(); start_timer(); - idct(data); + idct(&data[i*64]); stop_idct_timer(); - - start_timer(); - if (k < 4) - { - transfer_16to8add(dec->cur.y + (16*y + 4*(k&2))*stride + 16*x + 8*(k&1), data, stride); - } - else if (k == 4) - { - transfer_16to8add(dec->cur.u + 8*y*stride2 + 8*x, data, stride2); - } - else // k == 5 - { - transfer_16to8add(dec->cur.v + 8*y*stride2 + 8*x, data, stride2); - } - stop_transfer_timer(); } } -} + if (pMB->field_dct) + { + next_block = stride; + stride *= 2; + } + + start_timer(); + if (cbp & 32) + transfer_16to8add(pY_Cur, &data[0*64], stride); + if (cbp & 16) + transfer_16to8add(pY_Cur + 8, &data[1*64], stride); + if (cbp & 8) + transfer_16to8add(pY_Cur + next_block, &data[2*64], stride); + if (cbp & 4) + transfer_16to8add(pY_Cur + 8 + next_block, &data[3*64], stride); + if (cbp & 2) + transfer_16to8add(pU_Cur, &data[4*64], stride2); + if (cbp & 1) + transfer_16to8add(pV_Cur, &data[5*64], stride2); + stop_transfer_timer(); +} void decoder_iframe(DECODER * dec, Bitstream * bs, int quant, int intra_dc_threshold) { + uint32_t x, y; for (y = 0; y < dec->mb_height; y++) @@ -362,16 +402,23 @@ } } mb->quant = quant; - + + if (dec->interlacing) + { + mb->field_dct = BitstreamGetBit(bs); + DEBUG1("deci: field_dct: ", mb->field_dct); + } decoder_mbintra(dec, mb, x, y, acpred_flag, cbp, bs, quant, intra_dc_threshold); } } + } void get_motion_vector(DECODER *dec, Bitstream *bs, int x, int y, int k, VECTOR * mv, int fcode) { + int scale_fac = 1 << (fcode - 1); int high = (32 * scale_fac) - 1; int low = ((-32) * scale_fac); @@ -421,12 +468,13 @@ void decoder_pframe(DECODER * dec, Bitstream * bs, int rounding, int quant, int fcode, int intra_dc_threshold) { + uint32_t x, y; image_swap(&dec->cur, &dec->refn); start_timer(); - image_setedges(&dec->refn, dec->edged_width, dec->edged_height, dec->width, dec->height); + image_setedges(&dec->refn, dec->edged_width, dec->edged_height, dec->width, dec->height, dec->interlacing); stop_edges_timer(); for (y = 0; y < dec->mb_height; y++) @@ -478,13 +526,40 @@ } } mb->quant = quant; - - if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) + + if (dec->interlacing) { + mb->field_dct = BitstreamGetBit(bs); + DEBUG1("decp: field_dct: ", mb->field_dct); - get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode); - mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = mb->mvs[0].x; - mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = mb->mvs[0].y; + if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) + { + mb->field_pred = BitstreamGetBit(bs); + DEBUG1("decp: field_pred: ", mb->field_pred); + + if (mb->field_pred) + { + mb->field_for_top = BitstreamGetBit(bs); + DEBUG1("decp: field_for_top: ", mb->field_for_top); + mb->field_for_bot = BitstreamGetBit(bs); + DEBUG1("decp: field_for_bot: ", mb->field_for_bot); + } + } + } + + if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) + { + if (dec->interlacing && mb->field_pred) + { + get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode); + get_motion_vector(dec, bs, x, y, 0, &mb->mvs[1], fcode); + } + else + { + get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode); + mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = mb->mvs[0].x; + mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = mb->mvs[0].y; + } } else if (mb->mode == MODE_INTER4V /* || mb->mode == MODE_INTER4V_Q */) { @@ -515,28 +590,28 @@ start_timer(); transfer8x8_copy(dec->cur.y + (16*y)*dec->edged_width + (16*x), - dec->refn.y + (16*y)*dec->edged_width + (16*x), - dec->edged_width); + dec->refn.y + (16*y)*dec->edged_width + (16*x), + dec->edged_width); transfer8x8_copy(dec->cur.y + (16*y)*dec->edged_width + (16*x+8), - dec->refn.y + (16*y)*dec->edged_width + (16*x+8), - dec->edged_width); + dec->refn.y + (16*y)*dec->edged_width + (16*x+8), + dec->edged_width); transfer8x8_copy(dec->cur.y + (16*y+8)*dec->edged_width + (16*x), - dec->refn.y + (16*y+8)*dec->edged_width + (16*x), - dec->edged_width); + dec->refn.y + (16*y+8)*dec->edged_width + (16*x), + dec->edged_width); transfer8x8_copy(dec->cur.y + (16*y+8)*dec->edged_width + (16*x+8), - dec->refn.y + (16*y+8)*dec->edged_width + (16*x+8), - dec->edged_width); + dec->refn.y + (16*y+8)*dec->edged_width + (16*x+8), + dec->edged_width); transfer8x8_copy(dec->cur.u + (8*y)*dec->edged_width/2 + (8*x), - dec->refn.u + (8*y)*dec->edged_width/2 + (8*x), - dec->edged_width/2); + dec->refn.u + (8*y)*dec->edged_width/2 + (8*x), + dec->edged_width/2); transfer8x8_copy(dec->cur.v + (8*y)*dec->edged_width/2 + (8*x), - dec->refn.v + (8*y)*dec->edged_width/2 + (8*x), - dec->edged_width/2); + dec->refn.v + (8*y)*dec->edged_width/2 + (8*x), + dec->edged_width/2); stop_transfer_timer(); } @@ -546,6 +621,7 @@ int decoder_decode(DECODER * dec, XVID_DEC_FRAME * frame) { + Bitstream bs; uint32_t rounding; uint32_t quant; @@ -581,7 +657,7 @@ start_timer(); image_output(&dec->cur, dec->width, dec->height, dec->edged_width, - frame->image, frame->stride, frame->colorspace); + frame->image, frame->stride, frame->colorspace); stop_conv_timer(); emms(); @@ -589,4 +665,5 @@ stop_global_timer(); return XVID_ERR_OK; + }