Remove CONFIG_DIST_8X8 CONFIG_DIST_8X8 has not been used in the codebase. Removing it improves code readability. Change-Id: I670529e26fd0f57f53be9a8bc14714d5ecb21a4d
diff --git a/apps/aomenc.c b/apps/aomenc.c index cdc37ad..1f814bd 100644 --- a/apps/aomenc.c +++ b/apps/aomenc.c
@@ -593,11 +593,6 @@ ARG_DEF(NULL, "mv-cost-upd-freq", 1, "Update freq for mv costs" "0: SB, 1: SB Row per Tile, 2: Tile, 3: Off"); -#if CONFIG_DIST_8X8 -static const arg_def_t enable_dist_8x8 = - ARG_DEF(NULL, "enable-dist-8x8", 1, - "Enable dist-8x8 (0: false (default), 1: true)"); -#endif // CONFIG_DIST_8X8 static const arg_def_t num_tg = ARG_DEF( NULL, "num-tile-groups", 1, "Maximum number of tile groups, default is 1"); static const arg_def_t mtu_size = @@ -873,9 +868,6 @@ &coeff_cost_upd_freq, &mode_cost_upd_freq, &mv_cost_upd_freq, -#if CONFIG_DIST_8X8 - &enable_dist_8x8, -#endif &frame_parallel_decoding, &error_resilient_mode, &aq_mode, @@ -981,9 +973,6 @@ AV1E_SET_COEFF_COST_UPD_FREQ, AV1E_SET_MODE_COST_UPD_FREQ, AV1E_SET_MV_COST_UPD_FREQ, -#if CONFIG_DIST_8X8 - AV1E_SET_ENABLE_DIST_8X8, -#endif AV1E_SET_FRAME_PARALLEL_DECODING, AV1E_SET_ERROR_RESILIENT_MODE, AV1E_SET_AQ_MODE,
diff --git a/av1/av1_cx_iface.c b/av1/av1_cx_iface.c index 0930b72..d5cef3f 100644 --- a/av1/av1_cx_iface.c +++ b/av1/av1_cx_iface.c
@@ -64,9 +64,6 @@ unsigned int qm_v; unsigned int qm_min; unsigned int qm_max; -#if CONFIG_DIST_8X8 - unsigned int enable_dist_8x8; -#endif unsigned int num_tg; unsigned int mtu_size; @@ -189,11 +186,8 @@ DEFAULT_QM_V, // qm_v DEFAULT_QM_FIRST, // qm_min DEFAULT_QM_LAST, // qm_max -#if CONFIG_DIST_8X8 - 0, -#endif - 1, // max number of tile groups - 0, // mtu_size + 1, // max number of tile groups + 0, // mtu_size AOM_TIMING_UNSPECIFIED, // No picture timing signaling in bitstream 0, // frame_parallel_decoding_mode 1, // enable dual filter @@ -518,8 +512,6 @@ #if CONFIG_TUNE_VMAF RANGE_CHECK(extra_cfg, tuning, AOM_TUNE_PSNR, AOM_TUNE_VMAF_WITHOUT_PREPROCESSING); -#elif CONFIG_DIST_8X8 - RANGE_CHECK(extra_cfg, tuning, AOM_TUNE_PSNR, AOM_TUNE_DAALA_DIST); #else RANGE_CHECK(extra_cfg, tuning, AOM_TUNE_PSNR, AOM_TUNE_SSIM); #endif @@ -534,10 +526,6 @@ ERROR("Only --aq_mode=0 can be used with --lossless=1."); if (extra_cfg->enable_chroma_deltaq) ERROR("Only --enable_chroma_deltaq=0 can be used with --lossless=1."); -#if CONFIG_DIST_8X8 - if (extra_cfg->enable_dist_8x8) - ERROR("dist-8x8 cannot be used with lossless compression."); -#endif } if (cfg->rc_resize_mode != RESIZE_NONE && @@ -776,12 +764,6 @@ oxcf->coeff_cost_upd_freq = (COST_UPDATE_TYPE)extra_cfg->coeff_cost_upd_freq; oxcf->mode_cost_upd_freq = (COST_UPDATE_TYPE)extra_cfg->mode_cost_upd_freq; oxcf->mv_cost_upd_freq = (COST_UPDATE_TYPE)extra_cfg->mv_cost_upd_freq; -#if CONFIG_DIST_8X8 - oxcf->using_dist_8x8 = extra_cfg->enable_dist_8x8; - if (extra_cfg->tuning == AOM_TUNE_CDEF_DIST || - extra_cfg->tuning == AOM_TUNE_DAALA_DIST) - oxcf->using_dist_8x8 = 1; -#endif oxcf->num_tile_groups = extra_cfg->num_tg; // In large-scale tile encoding mode, num_tile_groups is always 1. if (cfg->large_scale_tile) oxcf->num_tile_groups = 1; @@ -1291,14 +1273,7 @@ extra_cfg.qm_max = CAST(AV1E_SET_QM_MAX, args); return update_extra_cfg(ctx, &extra_cfg); } -#if CONFIG_DIST_8X8 -static aom_codec_err_t ctrl_set_enable_dist_8x8(aom_codec_alg_priv_t *ctx, - va_list args) { - struct av1_extracfg extra_cfg = ctx->extra_cfg; - extra_cfg.enable_dist_8x8 = CAST(AV1E_SET_ENABLE_DIST_8X8, args); - return update_extra_cfg(ctx, &extra_cfg); -} -#endif + static aom_codec_err_t ctrl_set_num_tg(aom_codec_alg_priv_t *ctx, va_list args) { struct av1_extracfg extra_cfg = ctx->extra_cfg; @@ -2643,9 +2618,6 @@ { AV1E_SET_QM_V, ctrl_set_qm_v }, { AV1E_SET_QM_MIN, ctrl_set_qm_min }, { AV1E_SET_QM_MAX, ctrl_set_qm_max }, -#if CONFIG_DIST_8X8 - { AV1E_SET_ENABLE_DIST_8X8, ctrl_set_enable_dist_8x8 }, -#endif { AV1E_SET_NUM_TG, ctrl_set_num_tg }, { AV1E_SET_MTU, ctrl_set_mtu }, { AV1E_SET_TIMING_INFO_TYPE, ctrl_set_timing_info_type },
diff --git a/av1/encoder/block.h b/av1/encoder/block.h index b6a5ec4..0c439d4 100644 --- a/av1/encoder/block.h +++ b/av1/encoder/block.h
@@ -22,9 +22,6 @@ #endif #include "av1/encoder/hash.h" -#if CONFIG_DIST_8X8 -#include "aom/aomcx.h" -#endif #ifdef __cplusplus extern "C" { @@ -422,10 +419,6 @@ int use_default_intra_tx_type; // use default transform and skip transform type search for inter modes int use_default_inter_tx_type; -#if CONFIG_DIST_8X8 - int using_dist_8x8; - aom_tune_metric tune_metric; -#endif // CONFIG_DIST_8X8 int comp_idx_cost[COMP_INDEX_CONTEXTS][2]; int comp_group_idx_cost[COMP_GROUP_IDX_CONTEXTS][2]; int must_find_valid_partition;
diff --git a/av1/encoder/encodeframe.c b/av1/encoder/encodeframe.c index ca46152..3b06cdf 100644 --- a/av1/encoder/encodeframe.c +++ b/av1/encoder/encodeframe.c
@@ -2845,15 +2845,6 @@ unsigned int pb_simple_motion_pred_sse = UINT_MAX; (void)pb_simple_motion_pred_sse; -#if CONFIG_DIST_8X8 - if (x->using_dist_8x8) { - if (block_size_high[bsize] <= 8) partition_horz_allowed = 0; - if (block_size_wide[bsize] <= 8) partition_vert_allowed = 0; - if (block_size_high[bsize] <= 8 || block_size_wide[bsize] <= 8) - do_square_split = 0; - } -#endif - // PARTITION_NONE if (is_le_min_sq_part && has_rows && has_cols) partition_none_allowed = 1; int64_t part_none_rd = INT64_MAX; @@ -3276,15 +3267,6 @@ int vertab_partition_allowed = ext_partition_allowed & cpi->oxcf.enable_ab_partitions; -#if CONFIG_DIST_8X8 - if (x->using_dist_8x8) { - if (block_size_high[bsize] <= 8 || block_size_wide[bsize] <= 8) { - horzab_partition_allowed = 0; - vertab_partition_allowed = 0; - } - } -#endif - if (cpi->sf.part_sf.prune_ext_partition_types_search_level) { if (cpi->sf.part_sf.prune_ext_partition_types_search_level == 1) { // TODO(debargha,huisu@google.com): may need to tune the threshold for @@ -3571,15 +3553,6 @@ pb_source_variance, mi_row, mi_col); } -#if CONFIG_DIST_8X8 - if (x->using_dist_8x8) { - if (block_size_high[bsize] <= 16 || block_size_wide[bsize] <= 16) { - partition_horz4_allowed = 0; - partition_vert4_allowed = 0; - } - } -#endif - if (blksize < (min_partition_size << 2)) { partition_horz4_allowed = 0; partition_vert4_allowed = 0; @@ -5328,11 +5301,6 @@ RD_COUNTS *const rdc = &cpi->td.rd_counts; int i; -#if CONFIG_DIST_8X8 - x->using_dist_8x8 = cpi->oxcf.using_dist_8x8; - x->tune_metric = cpi->oxcf.tuning; -#endif - if (!cpi->sf.rt_sf.use_nonrd_pick_mode) { cm->setup_mi(cm); }
diff --git a/av1/encoder/encoder.h b/av1/encoder/encoder.h index fdb1165..997844c 100644 --- a/av1/encoder/encoder.h +++ b/av1/encoder/encoder.h
@@ -277,9 +277,6 @@ int qm_v; int qm_minlevel; int qm_maxlevel; -#if CONFIG_DIST_8X8 - int using_dist_8x8; -#endif unsigned int num_tile_groups; unsigned int mtu;
diff --git a/av1/encoder/pickcdef.c b/av1/encoder/pickcdef.c index 5f195c6..dd1f637 100644 --- a/av1/encoder/pickcdef.c +++ b/av1/encoder/pickcdef.c
@@ -174,36 +174,6 @@ memcpy(dst + r * dstride, base + r * sstride, hsize * sizeof(*base)); } -#if CONFIG_DIST_8X8 -static INLINE uint64_t dist_8x8_16bit(uint16_t *dst, int dstride, uint16_t *src, - int sstride, int coeff_shift) { - uint64_t svar = 0; - uint64_t dvar = 0; - uint64_t sum_s = 0; - uint64_t sum_d = 0; - uint64_t sum_s2 = 0; - uint64_t sum_d2 = 0; - uint64_t sum_sd = 0; - int i, j; - for (i = 0; i < 8; i++) { - for (j = 0; j < 8; j++) { - sum_s += src[i * sstride + j]; - sum_d += dst[i * dstride + j]; - sum_s2 += src[i * sstride + j] * src[i * sstride + j]; - sum_d2 += dst[i * dstride + j] * dst[i * dstride + j]; - sum_sd += src[i * sstride + j] * dst[i * dstride + j]; - } - } - /* Compute the variance -- the calculation cannot go negative. */ - svar = sum_s2 - ((sum_s * sum_s + 32) >> 6); - dvar = sum_d2 - ((sum_d * sum_d + 32) >> 6); - return (uint64_t)floor( - .5 + (sum_d2 + sum_s2 - 2 * sum_sd) * .5 * - (svar + dvar + (400 << 2 * coeff_shift)) / - (sqrt((20000 << 4 * coeff_shift) + svar * (double)dvar))); -} -#endif // CONFIG_DIST_8X8 - static INLINE uint64_t mse_8x8_16bit(uint16_t *dst, int dstride, uint16_t *src, int sstride) { uint64_t sum = 0; @@ -241,14 +211,8 @@ by = dlist[bi].by; bx = dlist[bi].bx; if (pli == 0) { -#if CONFIG_DIST_8X8 - sum += dist_8x8_16bit(&dst[(by << 3) * dstride + (bx << 3)], dstride, - &src[bi << (3 + 3)], 8, coeff_shift); -#else - sum += mse_8x8_16bit(&dst[(by << 3) * dstride + (bx << 3)], dstride, &src[bi << (3 + 3)], 8); -#endif // CONFIG_DIST_8X8 } else { sum += mse_8x8_16bit(&dst[(by << 3) * dstride + (bx << 3)], dstride,
diff --git a/av1/encoder/rdopt.c b/av1/encoder/rdopt.c index bd1a800..89dcc4c 100644 --- a/av1/encoder/rdopt.c +++ b/av1/encoder/rdopt.c
@@ -478,403 +478,6 @@ compare_rd_idx_pair); } -#if CONFIG_DIST_8X8 -static uint64_t cdef_dist_8x8_16bit(uint16_t *dst, int dstride, uint16_t *src, - int sstride, int coeff_shift) { - uint64_t svar = 0; - uint64_t dvar = 0; - uint64_t sum_s = 0; - uint64_t sum_d = 0; - uint64_t sum_s2 = 0; - uint64_t sum_d2 = 0; - uint64_t sum_sd = 0; - uint64_t dist = 0; - - int i, j; - for (i = 0; i < 8; i++) { - for (j = 0; j < 8; j++) { - sum_s += src[i * sstride + j]; - sum_d += dst[i * dstride + j]; - sum_s2 += src[i * sstride + j] * src[i * sstride + j]; - sum_d2 += dst[i * dstride + j] * dst[i * dstride + j]; - sum_sd += src[i * sstride + j] * dst[i * dstride + j]; - } - } - /* Compute the variance -- the calculation cannot go negative. */ - svar = sum_s2 - ((sum_s * sum_s + 32) >> 6); - dvar = sum_d2 - ((sum_d * sum_d + 32) >> 6); - - // Tuning of jm's original dering distortion metric used in CDEF tool, - // suggested by jm - const uint64_t a = 4; - const uint64_t b = 2; - const uint64_t c1 = (400 * a << 2 * coeff_shift); - const uint64_t c2 = (b * 20000 * a * a << 4 * coeff_shift); - - dist = (uint64_t)floor(.5 + (sum_d2 + sum_s2 - 2 * sum_sd) * .5 * - (svar + dvar + c1) / - (sqrt(svar * (double)dvar + c2))); - - // Calibrate dist to have similar rate for the same QP with MSE only - // distortion (as in master branch) - dist = (uint64_t)((float)dist * 0.75); - - return dist; -} - -static int od_compute_var_4x4(uint16_t *x, int stride) { - int sum; - int s2; - int i; - sum = 0; - s2 = 0; - for (i = 0; i < 4; i++) { - int j; - for (j = 0; j < 4; j++) { - int t; - - t = x[i * stride + j]; - sum += t; - s2 += t * t; - } - } - - return (s2 - (sum * sum >> 4)) >> 4; -} - -/* OD_DIST_LP_MID controls the frequency weighting filter used for computing - the distortion. For a value X, the filter is [1 X 1]/(X + 2) and - is applied both horizontally and vertically. For X=5, the filter is - a good approximation for the OD_QM8_Q4_HVS quantization matrix. */ -#define OD_DIST_LP_MID (5) -#define OD_DIST_LP_NORM (OD_DIST_LP_MID + 2) - -static double od_compute_dist_8x8(int use_activity_masking, uint16_t *x, - uint16_t *y, od_coeff *e_lp, int stride) { - double sum; - int min_var; - double mean_var; - double var_stat; - double activity; - double calibration; - int i; - int j; - double vardist; - - vardist = 0; - -#if 1 - min_var = INT_MAX; - mean_var = 0; - for (i = 0; i < 3; i++) { - for (j = 0; j < 3; j++) { - int varx; - int vary; - varx = od_compute_var_4x4(x + 2 * i * stride + 2 * j, stride); - vary = od_compute_var_4x4(y + 2 * i * stride + 2 * j, stride); - min_var = OD_MINI(min_var, varx); - mean_var += 1. / (1 + varx); - /* The cast to (double) is to avoid an overflow before the sqrt.*/ - vardist += varx - 2 * sqrt(varx * (double)vary) + vary; - } - } - /* We use a different variance statistic depending on whether activity - masking is used, since the harmonic mean appeared slightly worse with - masking off. The calibration constant just ensures that we preserve the - rate compared to activity=1. */ - if (use_activity_masking) { - calibration = 1.95; - var_stat = 9. / mean_var; - } else { - calibration = 1.62; - var_stat = min_var; - } - /* 1.62 is a calibration constant, 0.25 is a noise floor and 1/6 is the - activity masking constant. */ - activity = calibration * pow(.25 + var_stat, -1. / 6); -#else - activity = 1; -#endif // 1 - sum = 0; - for (i = 0; i < 8; i++) { - for (j = 0; j < 8; j++) - sum += e_lp[i * stride + j] * (double)e_lp[i * stride + j]; - } - /* Normalize the filter to unit DC response. */ - sum *= 1. / (OD_DIST_LP_NORM * OD_DIST_LP_NORM * OD_DIST_LP_NORM * - OD_DIST_LP_NORM); - return activity * activity * (sum + vardist); -} - -// Note : Inputs x and y are in a pixel domain -static double od_compute_dist_common(int activity_masking, uint16_t *x, - uint16_t *y, int bsize_w, int bsize_h, - int qindex, od_coeff *tmp, - od_coeff *e_lp) { - int i, j; - double sum = 0; - const int mid = OD_DIST_LP_MID; - - for (j = 0; j < bsize_w; j++) { - e_lp[j] = mid * tmp[j] + 2 * tmp[bsize_w + j]; - e_lp[(bsize_h - 1) * bsize_w + j] = mid * tmp[(bsize_h - 1) * bsize_w + j] + - 2 * tmp[(bsize_h - 2) * bsize_w + j]; - } - for (i = 1; i < bsize_h - 1; i++) { - for (j = 0; j < bsize_w; j++) { - e_lp[i * bsize_w + j] = mid * tmp[i * bsize_w + j] + - tmp[(i - 1) * bsize_w + j] + - tmp[(i + 1) * bsize_w + j]; - } - } - for (i = 0; i < bsize_h; i += 8) { - for (j = 0; j < bsize_w; j += 8) { - sum += od_compute_dist_8x8(activity_masking, &x[i * bsize_w + j], - &y[i * bsize_w + j], &e_lp[i * bsize_w + j], - bsize_w); - } - } - /* Scale according to linear regression against SSE, for 8x8 blocks. */ - if (activity_masking) { - sum *= 2.2 + (1.7 - 2.2) * (qindex - 99) / (210 - 99) + - (qindex < 99 ? 2.5 * (qindex - 99) / 99 * (qindex - 99) / 99 : 0); - } else { - sum *= qindex >= 128 - ? 1.4 + (0.9 - 1.4) * (qindex - 128) / (209 - 128) - : qindex <= 43 ? 1.5 + (2.0 - 1.5) * (qindex - 43) / (16 - 43) - : 1.5 + (1.4 - 1.5) * (qindex - 43) / (128 - 43); - } - - return sum; -} - -static double od_compute_dist(uint16_t *x, uint16_t *y, int bsize_w, - int bsize_h, int qindex) { - assert(bsize_w >= 8 && bsize_h >= 8); - - int activity_masking = 0; - - int i, j; - DECLARE_ALIGNED(16, od_coeff, e[MAX_SB_SQUARE]); - DECLARE_ALIGNED(16, od_coeff, tmp[MAX_SB_SQUARE]); - DECLARE_ALIGNED(16, od_coeff, e_lp[MAX_SB_SQUARE]); - for (i = 0; i < bsize_h; i++) { - for (j = 0; j < bsize_w; j++) { - e[i * bsize_w + j] = x[i * bsize_w + j] - y[i * bsize_w + j]; - } - } - int mid = OD_DIST_LP_MID; - for (i = 0; i < bsize_h; i++) { - tmp[i * bsize_w] = mid * e[i * bsize_w] + 2 * e[i * bsize_w + 1]; - tmp[i * bsize_w + bsize_w - 1] = - mid * e[i * bsize_w + bsize_w - 1] + 2 * e[i * bsize_w + bsize_w - 2]; - for (j = 1; j < bsize_w - 1; j++) { - tmp[i * bsize_w + j] = mid * e[i * bsize_w + j] + e[i * bsize_w + j - 1] + - e[i * bsize_w + j + 1]; - } - } - return od_compute_dist_common(activity_masking, x, y, bsize_w, bsize_h, - qindex, tmp, e_lp); -} - -static double od_compute_dist_diff(uint16_t *x, int16_t *e, int bsize_w, - int bsize_h, int qindex) { - assert(bsize_w >= 8 && bsize_h >= 8); - - int activity_masking = 0; - - DECLARE_ALIGNED(16, uint16_t, y[MAX_SB_SQUARE]); - DECLARE_ALIGNED(16, od_coeff, tmp[MAX_SB_SQUARE]); - DECLARE_ALIGNED(16, od_coeff, e_lp[MAX_SB_SQUARE]); - int i, j; - for (i = 0; i < bsize_h; i++) { - for (j = 0; j < bsize_w; j++) { - y[i * bsize_w + j] = x[i * bsize_w + j] - e[i * bsize_w + j]; - } - } - int mid = OD_DIST_LP_MID; - for (i = 0; i < bsize_h; i++) { - tmp[i * bsize_w] = mid * e[i * bsize_w] + 2 * e[i * bsize_w + 1]; - tmp[i * bsize_w + bsize_w - 1] = - mid * e[i * bsize_w + bsize_w - 1] + 2 * e[i * bsize_w + bsize_w - 2]; - for (j = 1; j < bsize_w - 1; j++) { - tmp[i * bsize_w + j] = mid * e[i * bsize_w + j] + e[i * bsize_w + j - 1] + - e[i * bsize_w + j + 1]; - } - } - return od_compute_dist_common(activity_masking, x, y, bsize_w, bsize_h, - qindex, tmp, e_lp); -} - -int64_t av1_dist_8x8(const AV1_COMP *const cpi, const MACROBLOCK *x, - const uint8_t *src, int src_stride, const uint8_t *dst, - int dst_stride, const BLOCK_SIZE tx_bsize, int bsw, - int bsh, int visible_w, int visible_h, int qindex) { - int64_t d = 0; - int i, j; - const MACROBLOCKD *xd = &x->e_mbd; - - DECLARE_ALIGNED(16, uint16_t, orig[MAX_SB_SQUARE]); - DECLARE_ALIGNED(16, uint16_t, rec[MAX_SB_SQUARE]); - - assert(bsw >= 8); - assert(bsh >= 8); - assert((bsw & 0x07) == 0); - assert((bsh & 0x07) == 0); - - if (x->tune_metric == AOM_TUNE_CDEF_DIST || - x->tune_metric == AOM_TUNE_DAALA_DIST) { - if (is_cur_buf_hbd(xd)) { - for (j = 0; j < bsh; j++) - for (i = 0; i < bsw; i++) - orig[j * bsw + i] = CONVERT_TO_SHORTPTR(src)[j * src_stride + i]; - - if ((bsw == visible_w) && (bsh == visible_h)) { - for (j = 0; j < bsh; j++) - for (i = 0; i < bsw; i++) - rec[j * bsw + i] = CONVERT_TO_SHORTPTR(dst)[j * dst_stride + i]; - } else { - for (j = 0; j < visible_h; j++) - for (i = 0; i < visible_w; i++) - rec[j * bsw + i] = CONVERT_TO_SHORTPTR(dst)[j * dst_stride + i]; - - if (visible_w < bsw) { - for (j = 0; j < bsh; j++) - for (i = visible_w; i < bsw; i++) - rec[j * bsw + i] = CONVERT_TO_SHORTPTR(src)[j * src_stride + i]; - } - - if (visible_h < bsh) { - for (j = visible_h; j < bsh; j++) - for (i = 0; i < bsw; i++) - rec[j * bsw + i] = CONVERT_TO_SHORTPTR(src)[j * src_stride + i]; - } - } - } else { - for (j = 0; j < bsh; j++) - for (i = 0; i < bsw; i++) orig[j * bsw + i] = src[j * src_stride + i]; - - if ((bsw == visible_w) && (bsh == visible_h)) { - for (j = 0; j < bsh; j++) - for (i = 0; i < bsw; i++) rec[j * bsw + i] = dst[j * dst_stride + i]; - } else { - for (j = 0; j < visible_h; j++) - for (i = 0; i < visible_w; i++) - rec[j * bsw + i] = dst[j * dst_stride + i]; - - if (visible_w < bsw) { - for (j = 0; j < bsh; j++) - for (i = visible_w; i < bsw; i++) - rec[j * bsw + i] = src[j * src_stride + i]; - } - - if (visible_h < bsh) { - for (j = visible_h; j < bsh; j++) - for (i = 0; i < bsw; i++) - rec[j * bsw + i] = src[j * src_stride + i]; - } - } - } - } - - if (x->tune_metric == AOM_TUNE_DAALA_DIST) { - d = (int64_t)od_compute_dist(orig, rec, bsw, bsh, qindex); - } else if (x->tune_metric == AOM_TUNE_CDEF_DIST) { - int coeff_shift = AOMMAX(xd->bd - 8, 0); - - for (i = 0; i < bsh; i += 8) { - for (j = 0; j < bsw; j += 8) { - d += cdef_dist_8x8_16bit(&rec[i * bsw + j], bsw, &orig[i * bsw + j], - bsw, coeff_shift); - } - } - if (is_cur_buf_hbd(xd)) d = ((uint64_t)d) >> 2 * coeff_shift; - } else { - // Otherwise, MSE by default - d = pixel_dist_visible_only(cpi, x, src, src_stride, dst, dst_stride, - tx_bsize, bsh, bsw, visible_h, visible_w); - } - - return d; -} - -static int64_t dist_8x8_diff(const MACROBLOCK *x, const uint8_t *src, - int src_stride, const int16_t *diff, - int diff_stride, int bsw, int bsh, int visible_w, - int visible_h, int qindex) { - int64_t d = 0; - int i, j; - const MACROBLOCKD *xd = &x->e_mbd; - - DECLARE_ALIGNED(16, uint16_t, orig[MAX_SB_SQUARE]); - DECLARE_ALIGNED(16, int16_t, diff16[MAX_SB_SQUARE]); - - assert(bsw >= 8); - assert(bsh >= 8); - assert((bsw & 0x07) == 0); - assert((bsh & 0x07) == 0); - - if (x->tune_metric == AOM_TUNE_CDEF_DIST || - x->tune_metric == AOM_TUNE_DAALA_DIST) { - if (is_cur_buf_hbd(xd)) { - for (j = 0; j < bsh; j++) - for (i = 0; i < bsw; i++) - orig[j * bsw + i] = CONVERT_TO_SHORTPTR(src)[j * src_stride + i]; - } else { - for (j = 0; j < bsh; j++) - for (i = 0; i < bsw; i++) orig[j * bsw + i] = src[j * src_stride + i]; - } - - if ((bsw == visible_w) && (bsh == visible_h)) { - for (j = 0; j < bsh; j++) - for (i = 0; i < bsw; i++) - diff16[j * bsw + i] = diff[j * diff_stride + i]; - } else { - for (j = 0; j < visible_h; j++) - for (i = 0; i < visible_w; i++) - diff16[j * bsw + i] = diff[j * diff_stride + i]; - - if (visible_w < bsw) { - for (j = 0; j < bsh; j++) - for (i = visible_w; i < bsw; i++) diff16[j * bsw + i] = 0; - } - - if (visible_h < bsh) { - for (j = visible_h; j < bsh; j++) - for (i = 0; i < bsw; i++) diff16[j * bsw + i] = 0; - } - } - } - - if (x->tune_metric == AOM_TUNE_DAALA_DIST) { - d = (int64_t)od_compute_dist_diff(orig, diff16, bsw, bsh, qindex); - } else if (x->tune_metric == AOM_TUNE_CDEF_DIST) { - int coeff_shift = AOMMAX(xd->bd - 8, 0); - DECLARE_ALIGNED(16, uint16_t, dst16[MAX_SB_SQUARE]); - - for (i = 0; i < bsh; i++) { - for (j = 0; j < bsw; j++) { - dst16[i * bsw + j] = orig[i * bsw + j] - diff16[i * bsw + j]; - } - } - - for (i = 0; i < bsh; i += 8) { - for (j = 0; j < bsw; j += 8) { - d += cdef_dist_8x8_16bit(&dst16[i * bsw + j], bsw, &orig[i * bsw + j], - bsw, coeff_shift); - } - } - // Don't scale 'd' for HBD since it will be done by caller side for diff - // input - } else { - // Otherwise, MSE by default - d = aom_sum_squares_2d_i16(diff, diff_stride, visible_w, visible_h); - } - - return d; -} -#endif // CONFIG_DIST_8X8 - // Similar to get_horver_correlation, but also takes into account first // row/column, when computing horizontal/vertical correlation. void av1_get_horver_correlation_full_c(const int16_t *diff, int stride,
diff --git a/av1/encoder/rdopt.h b/av1/encoder/rdopt.h index e6aa75c..3a59ea4 100644 --- a/av1/encoder/rdopt.h +++ b/av1/encoder/rdopt.h
@@ -42,13 +42,6 @@ int av1_count_colors_highbd(const uint8_t *src8, int stride, int rows, int cols, int bit_depth, int *val_count); -#if CONFIG_DIST_8X8 -int64_t av1_dist_8x8(const struct AV1_COMP *const cpi, const MACROBLOCK *x, - const uint8_t *src, int src_stride, const uint8_t *dst, - int dst_stride, const BLOCK_SIZE tx_bsize, int bsw, - int bsh, int visible_w, int visible_h, int qindex); -#endif - static INLINE int av1_cost_skip_txb(MACROBLOCK *x, const TXB_CTX *const txb_ctx, int plane, TX_SIZE tx_size) { const TX_SIZE txs_ctx = get_txsize_entropy_ctx(tx_size);
diff --git a/av1/encoder/speed_features.c b/av1/encoder/speed_features.c index 5088b55..d558343 100644 --- a/av1/encoder/speed_features.c +++ b/av1/encoder/speed_features.c
@@ -1256,11 +1256,6 @@ tx_size_search_methods[cpi->sf.winner_mode_sf.tx_size_search_level], sizeof(cpi->tx_size_search_methods)); -#if CONFIG_DIST_8X8 - if (sf->tx_domain_dist_level > 0) cpi->oxcf.using_dist_8x8 = 0; - - if (cpi->oxcf.using_dist_8x8) x->min_partition_size = BLOCK_8X8; -#endif // CONFIG_DIST_8X8 if (cpi->oxcf.row_mt == 1 && (cpi->oxcf.max_threads > 1)) { if (sf->inter_sf.inter_mode_rd_model_estimation == 1) { // Revert to type 2
diff --git a/av1/encoder/tx_search.c b/av1/encoder/tx_search.c index 8398917..02b7b64 100644 --- a/av1/encoder/tx_search.c +++ b/av1/encoder/tx_search.c
@@ -374,19 +374,7 @@ NULL, &visible_cols, &visible_rows); const int diff_stride = block_size_wide[plane_bsize]; const int16_t *diff = x->plane[plane].src_diff; -#if CONFIG_DIST_8X8 - int txb_height = block_size_high[tx_bsize]; - int txb_width = block_size_wide[tx_bsize]; - if (x->using_dist_8x8 && plane == 0) { - const int src_stride = x->plane[plane].src.stride; - const int src_idx = (blk_row * src_stride + blk_col) << MI_SIZE_LOG2; - const int diff_idx = (blk_row * diff_stride + blk_col) << MI_SIZE_LOG2; - const uint8_t *src = &x->plane[plane].src.buf[src_idx]; - return dist_8x8_diff(x, src, src_stride, diff + diff_idx, diff_stride, - txb_width, txb_height, visible_cols, visible_rows, - x->qindex); - } -#endif + diff += ((blk_row * diff_stride + blk_col) << MI_SIZE_LOG2); uint64_t sse = aom_sum_squares_2d_i16(diff, diff_stride, visible_cols, visible_rows); @@ -1199,13 +1187,6 @@ assert(visible_rows > 0); assert(visible_cols > 0); -#if CONFIG_DIST_8X8 - if (x->using_dist_8x8 && plane == 0) - return (unsigned)av1_dist_8x8(cpi, x, src, src_stride, dst, dst_stride, - tx_bsize, txb_cols, txb_rows, visible_cols, - visible_rows, x->qindex); -#endif // CONFIG_DIST_8X8 - unsigned sse = pixel_dist_visible_only(cpi, x, src, src_stride, dst, dst_stride, tx_bsize, txb_rows, txb_cols, visible_rows, visible_cols); @@ -2211,9 +2192,6 @@ // Therefore transform domain distortion is not valid for these // transform sizes. txsize_sqr_up_map[tx_size] != TX_64X64; -#if CONFIG_DIST_8X8 - if (x->using_dist_8x8) use_transform_domain_distortion = 0; -#endif int calc_pixel_domain_distortion_final = x->use_transform_domain_distortion == 1 && use_transform_domain_distortion && x->rd_model != LOW_TXFM_RD; @@ -2608,10 +2586,6 @@ const int try_no_split = cpi->oxcf.enable_tx64 || txsize_sqr_up_map[tx_size] != TX_64X64; int try_split = tx_size > TX_4X4 && depth < MAX_VARTX_DEPTH; -#if CONFIG_DIST_8X8 - if (x->using_dist_8x8) - try_split &= tx_size_wide[tx_size] >= 16 && tx_size_high[tx_size] >= 16; -#endif TxCandidateInfo no_split = { INT64_MAX, 0, TX_TYPES }; // TX no split @@ -2787,11 +2761,6 @@ int64_t rd[MAX_TX_DEPTH + 1] = { INT64_MAX, INT64_MAX, INT64_MAX }; for (int n = start_tx; depth <= MAX_TX_DEPTH; depth++, n = sub_tx_size_map[n]) { -#if CONFIG_DIST_8X8 - if (x->using_dist_8x8) { - if (tx_size_wide[n] < 8 || tx_size_high[n] < 8) continue; - } -#endif if (!cpi->oxcf.enable_tx64 && txsize_sqr_up_map[n] == TX_64X64) continue; RD_STATS this_rd_stats;