-static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n,
- int height, int y_offset, int list)
-{
- int raw_my = h->mv_cache[list][scan8[n]][1];
- int filter_height_up = (raw_my & 3) ? 2 : 0;
- int filter_height_down = (raw_my & 3) ? 3 : 0;
- int full_my = (raw_my >> 2) + y_offset;
- int top = full_my - filter_height_up;
- int bottom = full_my + filter_height_down + height;
-
- return FFMAX(abs(top), bottom);
-}
-
-static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n,
- int height, int y_offset, int list0,
- int list1, int *nrefs)
-{
- int my;
-
- y_offset += 16 * (h->mb_y >> MB_FIELD(h));
-
- if (list0) {
- int ref_n = h->ref_cache[0][scan8[n]];
- Picture *ref = &h->ref_list[0][ref_n];
-
- // Error resilience puts the current picture in the ref list.
- // Don't try to wait on these as it will cause a deadlock.
- // Fields can wait on each other, though.
- if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
- (ref->reference & 3) != h->picture_structure) {
- my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
- if (refs[0][ref_n] < 0)
- nrefs[0] += 1;
- refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
- }
- }
-
- if (list1) {
- int ref_n = h->ref_cache[1][scan8[n]];
- Picture *ref = &h->ref_list[1][ref_n];
-
- if (ref->tf.progress->data != h->cur_pic.tf.progress->data ||
- (ref->reference & 3) != h->picture_structure) {
- my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
- if (refs[1][ref_n] < 0)
- nrefs[1] += 1;
- refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
- }
- }
-}
-
-/**
- * Wait until all reference frames are available for MC operations.
- *
- * @param h the H264 context
- */
-static void await_references(H264Context *h)
-{
- const int mb_xy = h->mb_xy;
- const int mb_type = h->cur_pic.mb_type[mb_xy];
- int refs[2][48];
- int nrefs[2] = { 0 };
- int ref, list;
-
- memset(refs, -1, sizeof(refs));
-
- if (IS_16X16(mb_type)) {
- get_lowest_part_y(h, refs, 0, 16, 0,
- IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
- } else if (IS_16X8(mb_type)) {
- get_lowest_part_y(h, refs, 0, 8, 0,
- IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
- get_lowest_part_y(h, refs, 8, 8, 8,
- IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
- } else if (IS_8X16(mb_type)) {
- get_lowest_part_y(h, refs, 0, 16, 0,
- IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
- get_lowest_part_y(h, refs, 4, 16, 0,
- IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
- } else {
- int i;
-
- assert(IS_8X8(mb_type));
-
- for (i = 0; i < 4; i++) {
- const int sub_mb_type = h->sub_mb_type[i];
- const int n = 4 * i;
- int y_offset = (i & 2) << 2;
-
- if (IS_SUB_8X8(sub_mb_type)) {
- get_lowest_part_y(h, refs, n, 8, y_offset,
- IS_DIR(sub_mb_type, 0, 0),
- IS_DIR(sub_mb_type, 0, 1),
- nrefs);
- } else if (IS_SUB_8X4(sub_mb_type)) {
- get_lowest_part_y(h, refs, n, 4, y_offset,
- IS_DIR(sub_mb_type, 0, 0),
- IS_DIR(sub_mb_type, 0, 1),
- nrefs);
- get_lowest_part_y(h, refs, n + 2, 4, y_offset + 4,
- IS_DIR(sub_mb_type, 0, 0),
- IS_DIR(sub_mb_type, 0, 1),
- nrefs);
- } else if (IS_SUB_4X8(sub_mb_type)) {
- get_lowest_part_y(h, refs, n, 8, y_offset,
- IS_DIR(sub_mb_type, 0, 0),
- IS_DIR(sub_mb_type, 0, 1),
- nrefs);
- get_lowest_part_y(h, refs, n + 1, 8, y_offset,
- IS_DIR(sub_mb_type, 0, 0),
- IS_DIR(sub_mb_type, 0, 1),
- nrefs);
- } else {
- int j;
- assert(IS_SUB_4X4(sub_mb_type));
- for (j = 0; j < 4; j++) {
- int sub_y_offset = y_offset + 2 * (j & 2);
- get_lowest_part_y(h, refs, n + j, 4, sub_y_offset,
- IS_DIR(sub_mb_type, 0, 0),
- IS_DIR(sub_mb_type, 0, 1),
- nrefs);
- }
- }
- }
- }
-
- for (list = h->list_count - 1; list >= 0; list--)
- for (ref = 0; ref < 48 && nrefs[list]; ref++) {
- int row = refs[list][ref];
- if (row >= 0) {
- Picture *ref_pic = &h->ref_list[list][ref];
- int ref_field = ref_pic->reference - 1;
- int ref_field_picture = ref_pic->field_picture;
- int pic_height = 16 * h->mb_height >> ref_field_picture;
-
- row <<= MB_MBAFF(h);
- nrefs[list]--;
-
- if (!FIELD_PICTURE(h) && ref_field_picture) { // frame referencing two fields
- ff_thread_await_progress(&ref_pic->tf,
- FFMIN((row >> 1) - !(row & 1),
- pic_height - 1),
- 1);
- ff_thread_await_progress(&ref_pic->tf,
- FFMIN((row >> 1), pic_height - 1),
- 0);
- } else if (FIELD_PICTURE(h) && !ref_field_picture) { // field referencing one field of a frame
- ff_thread_await_progress(&ref_pic->tf,
- FFMIN(row * 2 + ref_field,
- pic_height - 1),
- 0);
- } else if (FIELD_PICTURE(h)) {
- ff_thread_await_progress(&ref_pic->tf,
- FFMIN(row, pic_height - 1),
- ref_field);
- } else {
- ff_thread_await_progress(&ref_pic->tf,
- FFMIN(row, pic_height - 1),
- 0);
- }
- }
- }
-}
-
-static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
- int n, int square, int height,
- int delta, int list,
- uint8_t *dest_y, uint8_t *dest_cb,
- uint8_t *dest_cr,
- int src_x_offset, int src_y_offset,
- qpel_mc_func *qpix_op,
- h264_chroma_mc_func chroma_op,
- int pixel_shift, int chroma_idc)
-{
- const int mx = h->mv_cache[list][scan8[n]][0] + src_x_offset * 8;
- int my = h->mv_cache[list][scan8[n]][1] + src_y_offset * 8;
- const int luma_xy = (mx & 3) + ((my & 3) << 2);
- ptrdiff_t offset = ((mx >> 2) << pixel_shift) + (my >> 2) * h->mb_linesize;
- uint8_t *src_y = pic->f.data[0] + offset;
- uint8_t *src_cb, *src_cr;
- int extra_width = 0;
- int extra_height = 0;
- int emu = 0;
- const int full_mx = mx >> 2;
- const int full_my = my >> 2;
- const int pic_width = 16 * h->mb_width;
- const int pic_height = 16 * h->mb_height >> MB_FIELD(h);
- int ysh;
-
- if (mx & 7)
- extra_width -= 3;
- if (my & 7)
- extra_height -= 3;
-
- if (full_mx < 0 - extra_width ||
- full_my < 0 - extra_height ||
- full_mx + 16 /*FIXME*/ > pic_width + extra_width ||
- full_my + 16 /*FIXME*/ > pic_height + extra_height) {
- h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
- src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
- h->mb_linesize,
- 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
- full_my - 2, pic_width, pic_height);
- src_y = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
- emu = 1;
- }
-
- qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); // FIXME try variable height perhaps?
- if (!square)
- qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
-
- if (CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY)
- return;
-
- if (chroma_idc == 3 /* yuv444 */) {
- src_cb = pic->f.data[1] + offset;
- if (emu) {
- h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
- src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
- h->mb_linesize,
- 16 + 5, 16 + 5 /*FIXME*/,
- full_mx - 2, full_my - 2,
- pic_width, pic_height);
- src_cb = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
- }
- qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps?
- if (!square)
- qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
-
- src_cr = pic->f.data[2] + offset;
- if (emu) {
- h->vdsp.emulated_edge_mc(h->edge_emu_buffer,
- src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
- h->mb_linesize,
- 16 + 5, 16 + 5 /*FIXME*/,
- full_mx - 2, full_my - 2,
- pic_width, pic_height);
- src_cr = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
- }
- qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps?
- if (!square)
- qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
- return;
- }
-
- ysh = 3 - (chroma_idc == 2 /* yuv422 */);
- if (chroma_idc == 1 /* yuv420 */ && MB_FIELD(h)) {
- // chroma offset when predicting from a field of opposite parity
- my += 2 * ((h->mb_y & 1) - (pic->reference - 1));
- emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
- }
-
- src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) +
- (my >> ysh) * h->mb_uvlinesize;
- src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) +
- (my >> ysh) * h->mb_uvlinesize;
-
- if (emu) {
- h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb, h->mb_uvlinesize,
- 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
- pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
- src_cb = h->edge_emu_buffer;
- }
- chroma_op(dest_cb, src_cb, h->mb_uvlinesize,
- height >> (chroma_idc == 1 /* yuv420 */),
- mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
-
- if (emu) {
- h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr, h->mb_uvlinesize,
- 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
- pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
- src_cr = h->edge_emu_buffer;
- }
- chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
- mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
-}
-
-static av_always_inline void mc_part_std(H264Context *h, int n, int square,
- int height, int delta,
- uint8_t *dest_y, uint8_t *dest_cb,
- uint8_t *dest_cr,
- int x_offset, int y_offset,
- qpel_mc_func *qpix_put,
- h264_chroma_mc_func chroma_put,
- qpel_mc_func *qpix_avg,
- h264_chroma_mc_func chroma_avg,
- int list0, int list1,
- int pixel_shift, int chroma_idc)
-{
- qpel_mc_func *qpix_op = qpix_put;
- h264_chroma_mc_func chroma_op = chroma_put;
-
- dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
- if (chroma_idc == 3 /* yuv444 */) {
- dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
- dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
- } else if (chroma_idc == 2 /* yuv422 */) {
- dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
- dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
- } else { /* yuv420 */
- dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
- dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
- }
- x_offset += 8 * h->mb_x;
- y_offset += 8 * (h->mb_y >> MB_FIELD(h));
-
- if (list0) {
- Picture *ref = &h->ref_list[0][h->ref_cache[0][scan8[n]]];
- mc_dir_part(h, ref, n, square, height, delta, 0,
- dest_y, dest_cb, dest_cr, x_offset, y_offset,
- qpix_op, chroma_op, pixel_shift, chroma_idc);
-
- qpix_op = qpix_avg;
- chroma_op = chroma_avg;
- }
-
- if (list1) {
- Picture *ref = &h->ref_list[1][h->ref_cache[1][scan8[n]]];
- mc_dir_part(h, ref, n, square, height, delta, 1,
- dest_y, dest_cb, dest_cr, x_offset, y_offset,
- qpix_op, chroma_op, pixel_shift, chroma_idc);
- }
-}
-
-static av_always_inline void mc_part_weighted(H264Context *h, int n, int square,
- int height, int delta,
- uint8_t *dest_y, uint8_t *dest_cb,
- uint8_t *dest_cr,
- int x_offset, int y_offset,
- qpel_mc_func *qpix_put,
- h264_chroma_mc_func chroma_put,
- h264_weight_func luma_weight_op,
- h264_weight_func chroma_weight_op,
- h264_biweight_func luma_weight_avg,
- h264_biweight_func chroma_weight_avg,
- int list0, int list1,
- int pixel_shift, int chroma_idc)
-{
- int chroma_height;
-
- dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
- if (chroma_idc == 3 /* yuv444 */) {
- chroma_height = height;
- chroma_weight_avg = luma_weight_avg;
- chroma_weight_op = luma_weight_op;
- dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
- dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
- } else if (chroma_idc == 2 /* yuv422 */) {
- chroma_height = height;
- dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
- dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
- } else { /* yuv420 */
- chroma_height = height >> 1;
- dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
- dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
- }
- x_offset += 8 * h->mb_x;
- y_offset += 8 * (h->mb_y >> MB_FIELD(h));
-
- if (list0 && list1) {
- /* don't optimize for luma-only case, since B-frames usually
- * use implicit weights => chroma too. */
- uint8_t *tmp_cb = h->bipred_scratchpad;
- uint8_t *tmp_cr = h->bipred_scratchpad + (16 << pixel_shift);
- uint8_t *tmp_y = h->bipred_scratchpad + 16 * h->mb_uvlinesize;
- int refn0 = h->ref_cache[0][scan8[n]];
- int refn1 = h->ref_cache[1][scan8[n]];
-
- mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
- dest_y, dest_cb, dest_cr,
- x_offset, y_offset, qpix_put, chroma_put,
- pixel_shift, chroma_idc);
- mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
- tmp_y, tmp_cb, tmp_cr,
- x_offset, y_offset, qpix_put, chroma_put,
- pixel_shift, chroma_idc);
-
- if (h->use_weight == 2) {
- int weight0 = h->implicit_weight[refn0][refn1][h->mb_y & 1];
- int weight1 = 64 - weight0;
- luma_weight_avg(dest_y, tmp_y, h->mb_linesize,
- height, 5, weight0, weight1, 0);
- chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
- chroma_height, 5, weight0, weight1, 0);
- chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
- chroma_height, 5, weight0, weight1, 0);
- } else {
- luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height,
- h->luma_log2_weight_denom,
- h->luma_weight[refn0][0][0],
- h->luma_weight[refn1][1][0],
- h->luma_weight[refn0][0][1] +
- h->luma_weight[refn1][1][1]);
- chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height,
- h->chroma_log2_weight_denom,
- h->chroma_weight[refn0][0][0][0],
- h->chroma_weight[refn1][1][0][0],
- h->chroma_weight[refn0][0][0][1] +
- h->chroma_weight[refn1][1][0][1]);
- chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height,
- h->chroma_log2_weight_denom,
- h->chroma_weight[refn0][0][1][0],
- h->chroma_weight[refn1][1][1][0],
- h->chroma_weight[refn0][0][1][1] +
- h->chroma_weight[refn1][1][1][1]);
- }
- } else {
- int list = list1 ? 1 : 0;
- int refn = h->ref_cache[list][scan8[n]];
- Picture *ref = &h->ref_list[list][refn];
- mc_dir_part(h, ref, n, square, height, delta, list,
- dest_y, dest_cb, dest_cr, x_offset, y_offset,
- qpix_put, chroma_put, pixel_shift, chroma_idc);
-
- luma_weight_op(dest_y, h->mb_linesize, height,
- h->luma_log2_weight_denom,
- h->luma_weight[refn][list][0],
- h->luma_weight[refn][list][1]);
- if (h->use_weight_chroma) {
- chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height,
- h->chroma_log2_weight_denom,
- h->chroma_weight[refn][list][0][0],
- h->chroma_weight[refn][list][0][1]);
- chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height,
- h->chroma_log2_weight_denom,
- h->chroma_weight[refn][list][1][0],
- h->chroma_weight[refn][list][1][1]);
- }
- }
-}
-
-static av_always_inline void prefetch_motion(H264Context *h, int list,
- int pixel_shift, int chroma_idc)
-{
- /* fetch pixels for estimated mv 4 macroblocks ahead
- * optimized for 64byte cache lines */
- const int refn = h->ref_cache[list][scan8[0]];
- if (refn >= 0) {
- const int mx = (h->mv_cache[list][scan8[0]][0] >> 2) + 16 * h->mb_x + 8;
- const int my = (h->mv_cache[list][scan8[0]][1] >> 2) + 16 * h->mb_y;
- uint8_t **src = h->ref_list[list][refn].f.data;
- int off = (mx << pixel_shift) +
- (my + (h->mb_x & 3) * 4) * h->mb_linesize +
- (64 << pixel_shift);
- h->vdsp.prefetch(src[0] + off, h->linesize, 4);
- if (chroma_idc == 3 /* yuv444 */) {
- h->vdsp.prefetch(src[1] + off, h->linesize, 4);
- h->vdsp.prefetch(src[2] + off, h->linesize, 4);
- } else {
- off = ((mx >> 1) << pixel_shift) +
- ((my >> 1) + (h->mb_x & 7)) * h->uvlinesize +
- (64 << pixel_shift);
- h->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
- }
- }
-}
-
-static void free_tables(H264Context *h, int free_rbsp)