diff options
Diffstat (limited to 'vp9/encoder')
-rw-r--r-- | vp9/encoder/vp9_asm_enc_offsets.c | 2 | ||||
-rw-r--r-- | vp9/encoder/vp9_encodeintra.c | 40 | ||||
-rw-r--r-- | vp9/encoder/vp9_encodeintra.h | 1 | ||||
-rw-r--r-- | vp9/encoder/vp9_encodemb.c | 7 | ||||
-rw-r--r-- | vp9/encoder/vp9_quantize.c | 37 | ||||
-rw-r--r-- | vp9/encoder/vp9_rdopt.c | 136 | ||||
-rw-r--r-- | vp9/encoder/vp9_tokenize.c | 4 | ||||
-rw-r--r-- | vp9/encoder/x86/vp9_encodeopt.asm | 114 | ||||
-rw-r--r-- | vp9/encoder/x86/vp9_x86_csystemdependent.c | 18 |
9 files changed, 156 insertions, 203 deletions
diff --git a/vp9/encoder/vp9_asm_enc_offsets.c b/vp9/encoder/vp9_asm_enc_offsets.c index e174a894a..1a770dcf7 100644 --- a/vp9/encoder/vp9_asm_enc_offsets.c +++ b/vp9/encoder/vp9_asm_enc_offsets.c @@ -29,9 +29,7 @@ DEFINE(vp9_block_zbin_extra, offsetof(BLOCK, zbin_extra)); DEFINE(vp9_block_zrun_zbin_boost, offsetof(BLOCK, zrun_zbin_boost)); DEFINE(vp9_block_quant_shift, offsetof(BLOCK, quant_shift)); -DEFINE(vp9_blockd_qcoeff, offsetof(BLOCKD, qcoeff)); DEFINE(vp9_blockd_dequant, offsetof(BLOCKD, dequant)); -DEFINE(vp9_blockd_dqcoeff, offsetof(BLOCKD, dqcoeff)); END diff --git a/vp9/encoder/vp9_encodeintra.c b/vp9/encoder/vp9_encodeintra.c index eddacb872..883038b7e 100644 --- a/vp9/encoder/vp9_encodeintra.c +++ b/vp9/encoder/vp9_encodeintra.c @@ -16,6 +16,8 @@ #include "vp9/common/vp9_invtrans.h" #include "vp9/encoder/vp9_encodeintra.h" +static void encode_intra4x4block(MACROBLOCK *x, int ib); + int vp9_encode_intra(VP9_COMP *cpi, MACROBLOCK *x, int use_16x16_pred) { MB_MODE_INFO * mbmi = &x->e_mbd.mode_info_context->mbmi; (void) cpi; @@ -31,18 +33,21 @@ int vp9_encode_intra(VP9_COMP *cpi, MACROBLOCK *x, int use_16x16_pred) { for (i = 0; i < 16; i++) { x->e_mbd.block[i].bmi.as_mode.first = B_DC_PRED; - vp9_encode_intra4x4block(x, i); + encode_intra4x4block(x, i); } } return vp9_get_mb_ss(x->src_diff); } -void vp9_encode_intra4x4block(MACROBLOCK *x, int ib) { +static void encode_intra4x4block(MACROBLOCK *x, int ib) { BLOCKD *b = &x->e_mbd.block[ib]; BLOCK *be = &x->block[ib]; + MACROBLOCKD * const xd = &x->e_mbd; TX_TYPE tx_type; + assert(ib < 16); + #if CONFIG_NEWBINTRAMODES b->bmi.as_mode.context = vp9_find_bpred_context(&x->e_mbd, b); #endif @@ -54,12 +59,14 @@ void vp9_encode_intra4x4block(MACROBLOCK *x, int ib) { if (tx_type != DCT_DCT) { vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type); vp9_ht_quantize_b_4x4(x, ib, tx_type); - vp9_short_iht4x4(b->dqcoeff, b->diff, 16, tx_type); + vp9_short_iht4x4(BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16), + b->diff, 16, tx_type); } else { x->fwd_txm4x4(be->src_diff, be->coeff, 32); x->quantize_b_4x4(x, ib); vp9_inverse_transform_b_4x4(&x->e_mbd, x->e_mbd.eobs[ib], - b->dqcoeff, b->diff, 32); + BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16), + b->diff, 32); } vp9_recon_b(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride); @@ -69,7 +76,7 @@ void vp9_encode_intra4x4mby(MACROBLOCK *mb) { int i; for (i = 0; i < 16; i++) - vp9_encode_intra4x4block(mb, i); + encode_intra4x4block(mb, i); } void vp9_encode_intra16x16mby(VP9_COMMON *const cm, MACROBLOCK *x) { @@ -151,41 +158,47 @@ void vp9_encode_intra8x8(MACROBLOCK *x, int ib) { if (xd->mode_info_context->mbmi.txfm_size == TX_8X8) { int idx = (ib & 0x02) ? (ib + 2) : ib; + int16_t * const dqcoeff = BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16); + assert(idx < 16); tx_type = get_tx_type_8x8(xd, ib); if (tx_type != DCT_DCT) { vp9_short_fht8x8(be->src_diff, (x->block + idx)->coeff, 16, tx_type); x->quantize_b_8x8(x, idx, tx_type); - vp9_short_iht8x8(xd->block[idx].dqcoeff, xd->block[ib].diff, + vp9_short_iht8x8(dqcoeff, xd->block[ib].diff, 16, tx_type); } else { x->fwd_txm8x8(be->src_diff, (x->block + idx)->coeff, 32); x->quantize_b_8x8(x, idx, DCT_DCT); - vp9_short_idct8x8(xd->block[idx].dqcoeff, xd->block[ib].diff, 32); + vp9_short_idct8x8(dqcoeff, xd->block[ib].diff, 32); } } else { for (i = 0; i < 4; i++) { + int idx = ib + iblock[i]; + int16_t * const dqcoeff = BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16); + + assert(idx < 16); b = &xd->block[ib + iblock[i]]; be = &x->block[ib + iblock[i]]; tx_type = get_tx_type_4x4(xd, ib + iblock[i]); if (tx_type != DCT_DCT) { vp9_short_fht4x4(be->src_diff, be->coeff, 16, tx_type); vp9_ht_quantize_b_4x4(x, ib + iblock[i], tx_type); - vp9_short_iht4x4(b->dqcoeff, b->diff, 16, tx_type); + vp9_short_iht4x4(dqcoeff, b->diff, 16, tx_type); } else if (!(i & 1) && get_tx_type_4x4(xd, ib + iblock[i] + 1) == DCT_DCT) { x->fwd_txm8x4(be->src_diff, be->coeff, 32); x->quantize_b_4x4_pair(x, ib + iblock[i], ib + iblock[i] + 1); vp9_inverse_transform_b_4x4(xd, xd->eobs[ib + iblock[i]], - b->dqcoeff, b->diff, 32); + dqcoeff, b->diff, 32); vp9_inverse_transform_b_4x4(xd, xd->eobs[ib + iblock[i] + 1], - (b + 1)->dqcoeff, (b + 1)->diff, 32); + dqcoeff + 16, (b + 1)->diff, 32); i++; } else { x->fwd_txm4x4(be->src_diff, be->coeff, 32); x->quantize_b_4x4(x, ib + iblock[i]); vp9_inverse_transform_b_4x4(xd, xd->eobs[ib + iblock[i]], - b->dqcoeff, b->diff, 32); + dqcoeff, b->diff, 32); } } } @@ -206,9 +219,12 @@ void vp9_encode_intra8x8mby(MACROBLOCK *x) { } static void encode_intra_uv4x4(MACROBLOCK *x, int ib, int mode) { + MACROBLOCKD * const xd = &x->e_mbd; BLOCKD *b = &x->e_mbd.block[ib]; BLOCK *be = &x->block[ib]; + int16_t * const dqcoeff = MB_SUBBLOCK_FIELD(xd, dqcoeff, ib); + assert(ib >= 16 && ib < 24); vp9_intra_uv4x4_predict(&x->e_mbd, b, mode, b->predictor); vp9_subtract_b(be, b, 8); @@ -216,7 +232,7 @@ static void encode_intra_uv4x4(MACROBLOCK *x, int ib, int mode) { x->fwd_txm4x4(be->src_diff, be->coeff, 16); x->quantize_b_4x4(x, ib); vp9_inverse_transform_b_4x4(&x->e_mbd, x->e_mbd.eobs[ib], - b->dqcoeff, b->diff, 16); + dqcoeff, b->diff, 16); vp9_recon_uv_b_c(b->predictor, b->diff, *(b->base_dst) + b->dst, b->dst_stride); diff --git a/vp9/encoder/vp9_encodeintra.h b/vp9/encoder/vp9_encodeintra.h index 0b19b5652..6576c94d2 100644 --- a/vp9/encoder/vp9_encodeintra.h +++ b/vp9/encoder/vp9_encodeintra.h @@ -17,7 +17,6 @@ int vp9_encode_intra(VP9_COMP *cpi, MACROBLOCK *x, int use_16x16_pred); void vp9_encode_intra16x16mby(VP9_COMMON *const cm, MACROBLOCK *x); void vp9_encode_intra16x16mbuv(VP9_COMMON *const cm, MACROBLOCK *x); void vp9_encode_intra4x4mby(MACROBLOCK *mb); -void vp9_encode_intra4x4block(MACROBLOCK *x, int ib); void vp9_encode_intra8x8mby(MACROBLOCK *x); void vp9_encode_intra8x8mbuv(MACROBLOCK *x); void vp9_encode_intra8x8(MACROBLOCK *x, int ib); diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c index 27015773f..a30268886 100644 --- a/vp9/encoder/vp9_encodemb.c +++ b/vp9/encoder/vp9_encodemb.c @@ -549,9 +549,10 @@ static void optimize_b(VP9_COMMON *const cm, MACROBLOCKD *const xd = &mb->e_mbd; vp9_token_state tokens[1025][2]; unsigned best_index[1025][2]; + const struct plane_block_idx pb_idx = plane_block_idx(xd, ib); const int16_t *coeff_ptr = mb->coeff + ib * 16; - int16_t *qcoeff_ptr = xd->qcoeff + ib * 16; - int16_t *dqcoeff_ptr = xd->dqcoeff + ib * 16; + int16_t *qcoeff_ptr; + int16_t *dqcoeff_ptr; int eob = xd->eobs[ib], final_eob, sz = 0; const int i0 = 0; int rc, x, next, i; @@ -582,6 +583,8 @@ static void optimize_b(VP9_COMMON *const cm, nzc0 = nzc1 = nzc; #endif + dqcoeff_ptr = BLOCK_OFFSET(xd->plane[pb_idx.plane].dqcoeff, pb_idx.block, 16); + qcoeff_ptr = BLOCK_OFFSET(xd->plane[pb_idx.plane].qcoeff, pb_idx.block, 16); switch (tx_size) { default: case TX_4X4: { diff --git a/vp9/encoder/vp9_quantize.c b/vp9/encoder/vp9_quantize.c index 881fce50f..826bee4c3 100644 --- a/vp9/encoder/vp9_quantize.c +++ b/vp9/encoder/vp9_quantize.c @@ -39,8 +39,9 @@ void vp9_ht_quantize_b_4x4(MACROBLOCK *mb, int b_idx, TX_TYPE tx_type) { int zbin; int x, y, z, sz; int16_t *coeff_ptr = mb->coeff + b_idx * 16; - int16_t *qcoeff_ptr = xd->qcoeff + b_idx * 16; - int16_t *dqcoeff_ptr = xd->dqcoeff + b_idx * 16; + // ht is luma-only + int16_t *qcoeff_ptr = BLOCK_OFFSET(xd->plane[0].qcoeff, b_idx, 16); + int16_t *dqcoeff_ptr = BLOCK_OFFSET(xd->plane[0].dqcoeff, b_idx, 16); int16_t *zbin_boost_ptr = b->zrun_zbin_boost; int16_t *zbin_ptr = b->zbin; int16_t *round_ptr = b->round; @@ -110,14 +111,17 @@ void vp9_ht_quantize_b_4x4(MACROBLOCK *mb, int b_idx, TX_TYPE tx_type) { void vp9_regular_quantize_b_4x4(MACROBLOCK *mb, int b_idx) { MACROBLOCKD *const xd = &mb->e_mbd; const int c_idx = plane_idx(xd, b_idx); + const struct plane_block_idx pb_idx = plane_block_idx(xd, b_idx); BLOCK *const b = &mb->block[c_idx]; BLOCKD *const d = &xd->block[c_idx]; int i, rc, eob; int zbin; int x, y, z, sz; int16_t *coeff_ptr = mb->coeff + b_idx * 16; - int16_t *qcoeff_ptr = xd->qcoeff + b_idx * 16; - int16_t *dqcoeff_ptr = xd->dqcoeff + b_idx * 16; + int16_t *qcoeff_ptr = BLOCK_OFFSET(xd->plane[pb_idx.plane].qcoeff, + pb_idx.block, 16); + int16_t *dqcoeff_ptr = BLOCK_OFFSET(xd->plane[pb_idx.plane].dqcoeff, + pb_idx.block, 16); int16_t *zbin_boost_ptr = b->zrun_zbin_boost; int16_t *zbin_ptr = b->zbin; int16_t *round_ptr = b->round; @@ -186,9 +190,13 @@ void vp9_quantize_mby_4x4(MACROBLOCK *x) { void vp9_quantize_mbuv_4x4(MACROBLOCK *x) { int i; + const MACROBLOCKD * const xd = &x->e_mbd; + const BLOCK_SIZE_TYPE real_sb_type = xd->mode_info_context->mbmi.sb_type; + xd->mode_info_context->mbmi.sb_type = BLOCK_SIZE_MB16X16; for (i = 16; i < 24; i++) x->quantize_b_4x4(x, i); + xd->mode_info_context->mbmi.sb_type = real_sb_type; } void vp9_quantize_mb_4x4(MACROBLOCK *x) { @@ -198,9 +206,12 @@ void vp9_quantize_mb_4x4(MACROBLOCK *x) { void vp9_regular_quantize_b_8x8(MACROBLOCK *mb, int b_idx, TX_TYPE tx_type) { MACROBLOCKD *const xd = &mb->e_mbd; - int16_t *qcoeff_ptr = xd->qcoeff + 16 * b_idx; - int16_t *dqcoeff_ptr = xd->dqcoeff + 16 * b_idx; + const struct plane_block_idx pb_idx = plane_block_idx(xd, b_idx); const int c_idx = plane_idx(xd, b_idx); + int16_t *qcoeff_ptr = BLOCK_OFFSET(xd->plane[pb_idx.plane].qcoeff, + pb_idx.block, 16); + int16_t *dqcoeff_ptr = BLOCK_OFFSET(xd->plane[pb_idx.plane].dqcoeff, + pb_idx.block, 16); BLOCK *const b = &mb->block[c_idx]; BLOCKD *const d = &xd->block[c_idx]; const int *pt_scan; @@ -323,6 +334,9 @@ void vp9_quantize_mby_8x8(MACROBLOCK *x) { void vp9_quantize_mbuv_8x8(MACROBLOCK *x) { int i; + const MACROBLOCKD * const xd = &x->e_mbd; + const BLOCK_SIZE_TYPE real_sb_type = xd->mode_info_context->mbmi.sb_type; + xd->mode_info_context->mbmi.sb_type = BLOCK_SIZE_MB16X16; #if CONFIG_CODE_NONZEROCOUNT for (i = 16; i < 24; i ++) { @@ -331,6 +345,7 @@ void vp9_quantize_mbuv_8x8(MACROBLOCK *x) { #endif for (i = 16; i < 24; i += 4) x->quantize_b_8x8(x, i, DCT_DCT); + xd->mode_info_context->mbmi.sb_type = real_sb_type; } void vp9_quantize_mb_8x8(MACROBLOCK *x) { @@ -418,6 +433,7 @@ static void quantize(int16_t *zbin_boost_orig_ptr, void vp9_regular_quantize_b_16x16(MACROBLOCK *mb, int b_idx, TX_TYPE tx_type) { MACROBLOCKD *const xd = &mb->e_mbd; const int c_idx = plane_idx(xd, b_idx); + const struct plane_block_idx pb_idx = plane_block_idx(xd, b_idx); BLOCK *const b = &mb->block[c_idx]; BLOCKD *const d = &xd->block[c_idx]; const int *pt_scan; @@ -438,8 +454,8 @@ void vp9_regular_quantize_b_16x16(MACROBLOCK *mb, int b_idx, TX_TYPE tx_type) { mb->coeff + 16 * b_idx, 256, b->skip_block, b->zbin, b->round, b->quant, b->quant_shift, - xd->qcoeff + 16 * b_idx, - xd->dqcoeff + 16 * b_idx, + BLOCK_OFFSET(xd->plane[pb_idx.plane].qcoeff, pb_idx.block, 16), + BLOCK_OFFSET(xd->plane[pb_idx.plane].dqcoeff, pb_idx.block, 16), d->dequant, b->zbin_extra, &xd->eobs[b_idx], @@ -452,6 +468,7 @@ void vp9_regular_quantize_b_16x16(MACROBLOCK *mb, int b_idx, TX_TYPE tx_type) { void vp9_regular_quantize_b_32x32(MACROBLOCK *mb, int b_idx) { MACROBLOCKD *const xd = &mb->e_mbd; const int c_idx = plane_idx(xd, b_idx); + const struct plane_block_idx pb_idx = plane_block_idx(xd, b_idx); BLOCK *const b = &mb->block[c_idx]; BLOCKD *const d = &xd->block[c_idx]; @@ -460,8 +477,8 @@ void vp9_regular_quantize_b_32x32(MACROBLOCK *mb, int b_idx) { 1024, b->skip_block, b->zbin, b->round, b->quant, b->quant_shift, - xd->qcoeff + b_idx * 16, - xd->dqcoeff + b_idx * 16, + BLOCK_OFFSET(xd->plane[pb_idx.plane].qcoeff, pb_idx.block, 16), + BLOCK_OFFSET(xd->plane[pb_idx.plane].dqcoeff, pb_idx.block, 16), d->dequant, b->zbin_extra, &xd->eobs[b_idx], diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c index 34adc9915..82c5b5bcd 100644 --- a/vp9/encoder/vp9_rdopt.c +++ b/vp9/encoder/vp9_rdopt.c @@ -348,35 +348,36 @@ int vp9_block_error_c(int16_t *coeff, int16_t *dqcoeff, int block_size) { } int vp9_mbblock_error_c(MACROBLOCK *mb) { + MACROBLOCKD * const xd = &mb->e_mbd; BLOCK *be; - BLOCKD *bd; - int i, j; - int berror, error = 0; + int i; + int error = 0; for (i = 0; i < 16; i++) { be = &mb->block[i]; - bd = &mb->e_mbd.block[i]; - berror = 0; - for (j = 0; j < 16; j++) { - int this_diff = be->coeff[j] - bd->dqcoeff[j]; - berror += this_diff * this_diff; - } - error += berror; + error += vp9_block_error(be->coeff, + BLOCK_OFFSET(xd->plane[0].dqcoeff, i, 16), 16); } return error; } int vp9_mbuverror_c(MACROBLOCK *mb) { + MACROBLOCKD * const xd = &mb->e_mbd; BLOCK *be; - BLOCKD *bd; int i, error = 0; - for (i = 16; i < 24; i++) { + for (i = 16; i < 20; i++) { be = &mb->block[i]; - bd = &mb->e_mbd.block[i]; - - error += vp9_block_error_c(be->coeff, bd->dqcoeff, 16); + error += vp9_block_error(be->coeff, + BLOCK_OFFSET(xd->plane[1].dqcoeff, i - 16, 16), + 16); + } + for (i = 20; i < 24; i++) { + be = &mb->block[i]; + error += vp9_block_error(be->coeff, + BLOCK_OFFSET(xd->plane[2].dqcoeff, i - 20, 16), + 16); } return error; @@ -438,7 +439,9 @@ static INLINE int cost_coeffs(VP9_COMMON *const cm, MACROBLOCK *mb, int c = 0; int cost = 0, pad; const int *scan, *nb; - const int16_t *qcoeff_ptr = xd->qcoeff + ib * 16; + const struct plane_block_idx pb_idx = plane_block_idx(xd, ib); + const int16_t *qcoeff_ptr = BLOCK_OFFSET(xd->plane[pb_idx.plane].qcoeff, + pb_idx.block, 16); const int ref = mbmi->ref_frame != INTRA_FRAME; unsigned int (*token_costs)[PREV_COEF_CONTEXTS][MAX_ENTROPY_TOKENS] = mb->token_costs[tx_size][type][ref]; @@ -858,6 +861,26 @@ static int vp9_sb_block_error_c(int16_t *coeff, int16_t *dqcoeff, return error > INT_MAX ? INT_MAX : (int)error; } +static int vp9_sb_uv_block_error_c(int16_t *coeff, + int16_t *dqcoeff0, int16_t *dqcoeff1, + int block_size, int shift) { + int i; + int64_t error = 0; + + for (i = 0; i < block_size / 2; i++) { + unsigned int this_diff = coeff[i] - dqcoeff0[i]; + error += this_diff * this_diff; + } + coeff += block_size / 2; + for (i = 0; i < block_size / 2; i++) { + unsigned int this_diff = coeff[i] - dqcoeff1[i]; + error += this_diff * this_diff; + } + error >>= shift; + + return error > INT_MAX ? INT_MAX : (int)error; +} + static int rdcost_sby_4x4(VP9_COMMON *const cm, MACROBLOCK *x) { int cost = 0, b; MACROBLOCKD *const xd = &x->e_mbd; @@ -884,7 +907,7 @@ static void super_block_yrd_4x4(VP9_COMMON *const cm, MACROBLOCK *x, vp9_transform_sby_4x4(x); vp9_quantize_sby_4x4(x); - *distortion = vp9_sb_block_error_c(x->coeff, xd->dqcoeff, 1024, 2); + *distortion = vp9_sb_block_error_c(x->coeff, xd->plane[0].dqcoeff, 1024, 2); *rate = rdcost_sby_4x4(cm, x); *skippable = vp9_sby_is_skippable_4x4(xd); } @@ -915,7 +938,7 @@ static void super_block_yrd_8x8(VP9_COMMON *const cm, MACROBLOCK *x, vp9_transform_sby_8x8(x); vp9_quantize_sby_8x8(x); - *distortion = vp9_sb_block_error_c(x->coeff, xd->dqcoeff, 1024, 2); + *distortion = vp9_sb_block_error_c(x->coeff, xd->plane[0].dqcoeff, 1024, 2); *rate = rdcost_sby_8x8(cm, x); *skippable = vp9_sby_is_skippable_8x8(xd); } @@ -946,7 +969,7 @@ static void super_block_yrd_16x16(VP9_COMMON *const cm, MACROBLOCK *x, vp9_transform_sby_16x16(x); vp9_quantize_sby_16x16(x); - *distortion = vp9_sb_block_error_c(x->coeff, xd->dqcoeff, 1024, 2); + *distortion = vp9_sb_block_error_c(x->coeff, xd->plane[0].dqcoeff, 1024, 2); *rate = rdcost_sby_16x16(cm, x); *skippable = vp9_sby_is_skippable_16x16(xd); } @@ -971,7 +994,7 @@ static void super_block_yrd_32x32(VP9_COMMON *const cm, MACROBLOCK *x, vp9_transform_sby_32x32(x); vp9_quantize_sby_32x32(x); - *distortion = vp9_sb_block_error_c(x->coeff, xd->dqcoeff, 1024, 0); + *distortion = vp9_sb_block_error_c(x->coeff, xd->plane[0].dqcoeff, 1024, 0); *rate = rdcost_sby_32x32(cm, x); *skippable = vp9_sby_is_skippable_32x32(xd); } @@ -1022,7 +1045,7 @@ static void super_block64_yrd_4x4(VP9_COMMON *const cm, MACROBLOCK *x, vp9_transform_sb64y_4x4(x); vp9_quantize_sb64y_4x4(x); - *distortion = vp9_sb_block_error_c(x->coeff, xd->dqcoeff, 4096, 2); + *distortion = vp9_sb_block_error_c(x->coeff, xd->plane[0].dqcoeff, 4096, 2); *rate = rdcost_sb64y_4x4(cm, x); *skippable = vp9_sb64y_is_skippable_4x4(xd); } @@ -1053,7 +1076,7 @@ static void super_block64_yrd_8x8(VP9_COMMON *const cm, MACROBLOCK *x, vp9_transform_sb64y_8x8(x); vp9_quantize_sb64y_8x8(x); - *distortion = vp9_sb_block_error_c(x->coeff, xd->dqcoeff, 4096, 2); + *distortion = vp9_sb_block_error_c(x->coeff, xd->plane[0].dqcoeff, 4096, 2); *rate = rdcost_sb64y_8x8(cm, x); *skippable = vp9_sb64y_is_skippable_8x8(xd); } @@ -1085,7 +1108,7 @@ static void super_block64_yrd_16x16(VP9_COMMON *const cm, MACROBLOCK *x, vp9_transform_sb64y_16x16(x); vp9_quantize_sb64y_16x16(x); - *distortion = vp9_sb_block_error_c(x->coeff, xd->dqcoeff, 4096, 2); + *distortion = vp9_sb_block_error_c(x->coeff, xd->plane[0].dqcoeff, 4096, 2); *rate = rdcost_sb64y_16x16(cm, x); *skippable = vp9_sb64y_is_skippable_16x16(xd); } @@ -1117,7 +1140,7 @@ static void super_block64_yrd_32x32(VP9_COMMON *const cm, MACROBLOCK *x, vp9_transform_sb64y_32x32(x); vp9_quantize_sb64y_32x32(x); - *distortion = vp9_sb_block_error_c(x->coeff, xd->dqcoeff, 4096, 0); + *distortion = vp9_sb_block_error_c(x->coeff, xd->plane[0].dqcoeff, 4096, 0); *rate = rdcost_sb64y_32x32(cm, x); *skippable = vp9_sb64y_is_skippable_32x32(xd); } @@ -1163,8 +1186,8 @@ static void copy_predictor_8x8(uint8_t *dst, const uint8_t *predictor) { d[29] = p[29]; } -static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, BLOCK *be, - BLOCKD *b, B_PREDICTION_MODE *best_mode, +static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib, + B_PREDICTION_MODE *best_mode, int *bmode_costs, ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l, int *bestrate, int *bestratey, @@ -1175,6 +1198,8 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, BLOCK *be, int rate = 0; int distortion; VP9_COMMON *const cm = &cpi->common; + BLOCK *be = x->block + ib; + BLOCKD *b = xd->block + ib; ENTROPY_CONTEXT ta = *a, tempa = *a; ENTROPY_CONTEXT tl = *l, templ = *l; @@ -1188,6 +1213,7 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, BLOCK *be, DECLARE_ALIGNED_ARRAY(16, uint8_t, best_predictor, 16 * 4); DECLARE_ALIGNED_ARRAY(16, int16_t, best_dqcoeff, 16); + assert(ib < 16); #if CONFIG_NEWBINTRAMODES b->bmi.as_mode.context = vp9_find_bpred_context(xd, b); #endif @@ -1233,7 +1259,9 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, BLOCK *be, ratey = cost_coeffs(cm, x, b - xd->block, PLANE_TYPE_Y_WITH_DC, &tempa, &templ, TX_4X4); rate += ratey; - distortion = vp9_block_error(be->coeff, b->dqcoeff, 16) >> 2; + distortion = vp9_block_error(be->coeff, + BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16), + 16) >> 2; this_rd = RDCOST(x->rdmult, x->rddiv, rate, distortion); @@ -1247,7 +1275,7 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, BLOCK *be, *a = tempa; *l = templ; copy_predictor(best_predictor, b->predictor); - vpx_memcpy(best_dqcoeff, b->dqcoeff, 32); + vpx_memcpy(best_dqcoeff, BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16), 32); } } b->bmi.as_mode.first = (B_PREDICTION_MODE)(*best_mode); @@ -1304,7 +1332,7 @@ static int64_t rd_pick_intra4x4mby_modes(VP9_COMP *cpi, MACROBLOCK *mb, #endif total_rd += rd_pick_intra4x4block( - cpi, mb, mb->block + i, xd->block + i, &best_mode, + cpi, mb, i, &best_mode, bmode_costs, ta + vp9_block2above[TX_4X4][i], tl + vp9_block2left[TX_4X4][i], &r, &ry, &d); @@ -1504,6 +1532,7 @@ static int64_t rd_pick_intra8x8block(VP9_COMP *cpi, MACROBLOCK *x, int ib, // note the input and output index mapping int idx = (ib & 0x02) ? (ib + 2) : ib; + assert(ib < 16); for (mode = DC_PRED; mode <= TM_PRED; mode++) { int64_t this_rd; int rate_t = 0; @@ -1526,7 +1555,7 @@ static int64_t rd_pick_intra8x8block(VP9_COMP *cpi, MACROBLOCK *x, int ib, // compute quantization mse of 8x8 block distortion = vp9_block_error_c((x->block + idx)->coeff, - (xd->block + idx)->dqcoeff, 64); + BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16), 64); vpx_memcpy(&ta, a, sizeof(ENTROPY_CONTEXT_PLANES)); vpx_memcpy(&tl, l, sizeof(ENTROPY_CONTEXT_PLANES)); @@ -1569,7 +1598,9 @@ static int64_t rd_pick_intra8x8block(VP9_COMP *cpi, MACROBLOCK *x, int ib, x->fwd_txm4x4(be->src_diff, be->coeff, 32); x->quantize_b_4x4(x, ib + iblock[i]); } - distortion += vp9_block_error_c(be->coeff, b->dqcoeff, 16 << do_two); + distortion += vp9_block_error_c(be->coeff, + BLOCK_OFFSET(xd->plane[0].dqcoeff, ib + iblock[i], 16), + 16 << do_two); rate_t += cost_coeffs(cm, x, ib + iblock[i], PLANE_TYPE_Y_WITH_DC, i&1 ? ta1 : ta0, i&2 ? tl1 : tl0, TX_4X4); @@ -1598,8 +1629,10 @@ static int64_t rd_pick_intra8x8block(VP9_COMP *cpi, MACROBLOCK *x, int ib, best_rd = this_rd; *best_mode = mode; copy_predictor_8x8(best_predictor, b->predictor); - vpx_memcpy(best_dqcoeff, b->dqcoeff, 64); - vpx_memcpy(best_dqcoeff + 32, b->dqcoeff + 64, 64); + vpx_memcpy(best_dqcoeff, + BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16), 64); + vpx_memcpy(best_dqcoeff + 32, + BLOCK_OFFSET(xd->plane[0].dqcoeff, ib, 16) + 64, 64); } } b->bmi.as_mode.first = (*best_mode); @@ -1742,6 +1775,8 @@ static int rd_cost_mbuv_4x4(VP9_COMMON *const cm, MACROBLOCK *mb, int backup) { MACROBLOCKD *xd = &mb->e_mbd; ENTROPY_CONTEXT_PLANES t_above, t_left; ENTROPY_CONTEXT *ta, *tl; + const BLOCK_SIZE_TYPE real_sb_type = xd->mode_info_context->mbmi.sb_type; + xd->mode_info_context->mbmi.sb_type = BLOCK_SIZE_MB16X16; if (backup) { vpx_memcpy(&t_above, xd->above_context, sizeof(ENTROPY_CONTEXT_PLANES)); @@ -1760,6 +1795,7 @@ static int rd_cost_mbuv_4x4(VP9_COMMON *const cm, MACROBLOCK *mb, int backup) { tl + vp9_block2left[TX_4X4][b], TX_4X4); + xd->mode_info_context->mbmi.sb_type = real_sb_type; return cost; } @@ -1783,6 +1819,8 @@ static int rd_cost_mbuv_8x8(VP9_COMMON *const cm, MACROBLOCK *mb, int backup) { MACROBLOCKD *xd = &mb->e_mbd; ENTROPY_CONTEXT_PLANES t_above, t_left; ENTROPY_CONTEXT *ta, *tl; + const BLOCK_SIZE_TYPE real_sb_type = xd->mode_info_context->mbmi.sb_type; + xd->mode_info_context->mbmi.sb_type = BLOCK_SIZE_MB16X16; if (backup) { vpx_memcpy(&t_above, xd->above_context, sizeof(ENTROPY_CONTEXT_PLANES)); @@ -1800,6 +1838,7 @@ static int rd_cost_mbuv_8x8(VP9_COMMON *const cm, MACROBLOCK *mb, int backup) { ta + vp9_block2above[TX_8X8][b], tl + vp9_block2left[TX_8X8][b], TX_8X8); + xd->mode_info_context->mbmi.sb_type = real_sb_type; return cost; } @@ -1851,8 +1890,9 @@ static void rd_inter32x32_uv_16x16(VP9_COMMON *const cm, MACROBLOCK *x, vp9_quantize_sbuv_16x16(x); *rate = rd_cost_sbuv_16x16(cm, x, backup); - *distortion = vp9_sb_block_error_c(x->coeff + 1024, - xd->dqcoeff + 1024, 512, 2); + *distortion = vp9_sb_uv_block_error_c(x->coeff + 1024, + xd->plane[1].dqcoeff, + xd->plane[2].dqcoeff, 512, 2); *skip = vp9_sbuv_is_skippable_16x16(xd); } @@ -2127,8 +2167,9 @@ static void rd_inter64x64_uv_32x32(VP9_COMMON *const cm, MACROBLOCK *x, vp9_quantize_sb64uv_32x32(x); *rate = rd_cost_sb64uv_32x32(cm, x, backup); - *distortion = vp9_sb_block_error_c(x->coeff + 4096, - xd->dqcoeff + 4096, 2048, 0); + *distortion = vp9_sb_uv_block_error_c(x->coeff + 4096, + xd->plane[1].dqcoeff, + xd->plane[2].dqcoeff, 2048, 0); *skip = vp9_sb64uv_is_skippable_32x32(xd); } @@ -2466,7 +2507,8 @@ static int64_t encode_inter_mb_segment(VP9_COMMON *const cm, vp9_subtract_b(be, bd, 16); x->fwd_txm4x4(be->src_diff, be->coeff, 32); x->quantize_b_4x4(x, i); - thisdistortion = vp9_block_error(be->coeff, bd->dqcoeff, 16); + thisdistortion = vp9_block_error(be->coeff, + BLOCK_OFFSET(xd->plane[0].dqcoeff, i, 16), 16); *distortion += thisdistortion; *labelyrate += cost_coeffs(cm, x, i, PLANE_TYPE_Y_WITH_DC, ta + vp9_block2above[TX_4X4][i], @@ -2508,11 +2550,12 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm, const int use_second_ref = xd->mode_info_context->mbmi.second_ref_frame > 0; int which_mv; - int idx = (ib & 8) + ((ib & 2) << 1); - BLOCKD *bd = &xd->block[ib], *bd2 = &xd->block[idx]; + const int idx = (ib & 8) + ((ib & 2) << 1); + BLOCKD *bd = &xd->block[ib]; BLOCK *be = &x->block[ib], *be2 = &x->block[idx]; int thisdistortion; + assert(idx < 16); for (which_mv = 0; which_mv < 1 + use_second_ref; ++which_mv) { uint8_t **base_pre = which_mv ? bd->base_second_pre : bd->base_pre; @@ -2532,7 +2575,8 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm, if (otherrd) { x->fwd_txm8x8(be->src_diff, be2->coeff, 32); x->quantize_b_8x8(x, idx, DCT_DCT); - thisdistortion = vp9_block_error_c(be2->coeff, bd2->dqcoeff, 64); + thisdistortion = vp9_block_error_c(be2->coeff, + BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16), 64); otherdist += thisdistortion; xd->mode_info_context->mbmi.txfm_size = TX_8X8; othercost += cost_coeffs(cm, x, idx, PLANE_TYPE_Y_WITH_DC, @@ -2546,7 +2590,8 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm, be = &x->block[ib + iblock[j]]; x->fwd_txm8x4(be->src_diff, be->coeff, 32); x->quantize_b_4x4_pair(x, ib + iblock[j], ib + iblock[j] + 1); - thisdistortion = vp9_block_error_c(be->coeff, bd->dqcoeff, 32); + thisdistortion = vp9_block_error_c(be->coeff, + BLOCK_OFFSET(xd->plane[0].dqcoeff, ib + iblock[j], 16), 32); *distortion += thisdistortion; *labelyrate += cost_coeffs(cm, x, ib + iblock[j], PLANE_TYPE_Y_WITH_DC, @@ -2563,11 +2608,11 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm, } else /* 8x8 */ { if (otherrd) { for (j = 0; j < 4; j += 2) { - BLOCKD *bd = &xd->block[ib + iblock[j]]; BLOCK *be = &x->block[ib + iblock[j]]; x->fwd_txm8x4(be->src_diff, be->coeff, 32); x->quantize_b_4x4_pair(x, ib + iblock[j], ib + iblock[j] + 1); - thisdistortion = vp9_block_error_c(be->coeff, bd->dqcoeff, 32); + thisdistortion = vp9_block_error_c(be->coeff, + BLOCK_OFFSET(xd->plane[0].dqcoeff, ib + iblock[j], 16), 32); otherdist += thisdistortion; xd->mode_info_context->mbmi.txfm_size = TX_4X4; othercost += @@ -2586,7 +2631,8 @@ static int64_t encode_inter_mb_segment_8x8(VP9_COMMON *const cm, } x->fwd_txm8x8(be->src_diff, be2->coeff, 32); x->quantize_b_8x8(x, idx, DCT_DCT); - thisdistortion = vp9_block_error_c(be2->coeff, bd2->dqcoeff, 64); + thisdistortion = vp9_block_error_c(be2->coeff, + BLOCK_OFFSET(xd->plane[0].dqcoeff, idx, 16), 64); *distortion += thisdistortion; *labelyrate += cost_coeffs(cm, x, idx, PLANE_TYPE_Y_WITH_DC, ta + vp9_block2above[TX_8X8][idx], diff --git a/vp9/encoder/vp9_tokenize.c b/vp9/encoder/vp9_tokenize.c index 8f9e9da69..ab286fd8a 100644 --- a/vp9/encoder/vp9_tokenize.c +++ b/vp9/encoder/vp9_tokenize.c @@ -123,7 +123,9 @@ static void tokenize_b(VP9_COMP *cpi, int c = 0; const int eob = xd->eobs[ib]; /* one beyond last nonzero coeff */ TOKENEXTRA *t = *tp; /* store tokens starting here */ - int16_t *qcoeff_ptr = xd->qcoeff + 16 * ib; + const struct plane_block_idx pb_idx = plane_block_idx(xd, ib); + const int16_t *qcoeff_ptr = BLOCK_OFFSET(xd->plane[pb_idx.plane].qcoeff, + pb_idx.block, 16); int seg_eob, default_eob, pad; const int segment_id = mbmi->segment_id; const BLOCK_SIZE_TYPE sb_type = mbmi->sb_type; diff --git a/vp9/encoder/x86/vp9_encodeopt.asm b/vp9/encoder/x86/vp9_encodeopt.asm index 90c793d4f..51314a7a8 100644 --- a/vp9/encoder/x86/vp9_encodeopt.asm +++ b/vp9/encoder/x86/vp9_encodeopt.asm @@ -260,117 +260,3 @@ sym(vp9_mbblock_error_xmm_impl): UNSHADOW_ARGS pop rbp ret - - -;int vp9_mbuverror_mmx_impl(short *s_ptr, short *d_ptr); -global sym(vp9_mbuverror_mmx_impl) PRIVATE -sym(vp9_mbuverror_mmx_impl): - push rbp - mov rbp, rsp - SHADOW_ARGS_TO_STACK 2 - push rsi - push rdi - ; end prolog - - - mov rsi, arg(0) ;s_ptr - mov rdi, arg(1) ;d_ptr - - mov rcx, 16 - pxor mm7, mm7 - -.mbuverror_loop_mmx: - - movq mm1, [rsi] - movq mm2, [rdi] - - psubw mm1, mm2 - pmaddwd mm1, mm1 - - - movq mm3, [rsi+8] - movq mm4, [rdi+8] - - psubw mm3, mm4 - pmaddwd mm3, mm3 - - - paddd mm7, mm1 - paddd mm7, mm3 - - - add rsi, 16 - add rdi, 16 - - dec rcx - jnz .mbuverror_loop_mmx - - movq mm0, mm7 - psrlq mm7, 32 - - paddd mm0, mm7 - movq rax, mm0 - - pop rdi - pop rsi - ; begin epilog - UNSHADOW_ARGS - pop rbp - ret - - -;int vp9_mbuverror_xmm_impl(short *s_ptr, short *d_ptr); -global sym(vp9_mbuverror_xmm_impl) PRIVATE -sym(vp9_mbuverror_xmm_impl): - push rbp - mov rbp, rsp - SHADOW_ARGS_TO_STACK 2 - push rsi - push rdi - ; end prolog - - - mov rsi, arg(0) ;s_ptr - mov rdi, arg(1) ;d_ptr - - mov rcx, 16 - pxor xmm3, xmm3 - -.mbuverror_loop: - - movdqa xmm1, [rsi] - movdqa xmm2, [rdi] - - psubw xmm1, xmm2 - pmaddwd xmm1, xmm1 - - paddd xmm3, xmm1 - - add rsi, 16 - add rdi, 16 - - dec rcx - jnz .mbuverror_loop - - pxor xmm0, xmm0 - movdqa xmm1, xmm3 - - movdqa xmm2, xmm1 - punpckldq xmm1, xmm0 - - punpckhdq xmm2, xmm0 - paddd xmm1, xmm2 - - movdqa xmm2, xmm1 - - psrldq xmm1, 8 - paddd xmm1, xmm2 - - movq rax, xmm1 - - pop rdi - pop rsi - ; begin epilog - UNSHADOW_ARGS - pop rbp - ret diff --git a/vp9/encoder/x86/vp9_x86_csystemdependent.c b/vp9/encoder/x86/vp9_x86_csystemdependent.c index 2bf32c569..9557af119 100644 --- a/vp9/encoder/x86/vp9_x86_csystemdependent.c +++ b/vp9/encoder/x86/vp9_x86_csystemdependent.c @@ -26,17 +26,10 @@ void vp9_short_fdct8x4_mmx(short *input, short *output, int pitch) { int vp9_mbblock_error_mmx_impl(short *coeff_ptr, short *dcoef_ptr); int vp9_mbblock_error_mmx(MACROBLOCK *mb) { short *coeff_ptr = mb->block[0].coeff; - short *dcoef_ptr = mb->e_mbd.block[0].dqcoeff; + short *dcoef_ptr = mb->e_mbd.plane[0].dqcoeff; return vp9_mbblock_error_mmx_impl(coeff_ptr, dcoef_ptr); } -int vp9_mbuverror_mmx_impl(short *s_ptr, short *d_ptr); -int vp9_mbuverror_mmx(MACROBLOCK *mb) { - short *s_ptr = &mb->coeff[256]; - short *d_ptr = &mb->e_mbd.dqcoeff[256]; - return vp9_mbuverror_mmx_impl(s_ptr, d_ptr); -} - void vp9_subtract_b_mmx_impl(unsigned char *z, int src_stride, short *diff, unsigned char *predictor, int pitch); @@ -54,17 +47,10 @@ void vp9_subtract_b_mmx(BLOCK *be, BLOCKD *bd, int pitch) { int vp9_mbblock_error_xmm_impl(short *coeff_ptr, short *dcoef_ptr); int vp9_mbblock_error_xmm(MACROBLOCK *mb) { short *coeff_ptr = mb->block[0].coeff; - short *dcoef_ptr = mb->e_mbd.block[0].dqcoeff; + short *dcoef_ptr = mb->e_mbd.plane[0].dqcoeff; return vp9_mbblock_error_xmm_impl(coeff_ptr, dcoef_ptr); } -int vp9_mbuverror_xmm_impl(short *s_ptr, short *d_ptr); -int vp9_mbuverror_xmm(MACROBLOCK *mb) { - short *s_ptr = &mb->coeff[256]; - short *d_ptr = &mb->e_mbd.dqcoeff[256]; - return vp9_mbuverror_xmm_impl(s_ptr, d_ptr); -} - void vp9_subtract_b_sse2_impl(unsigned char *z, int src_stride, short *diff, unsigned char *predictor, int pitch); |