void *aom_calloc(size_t num, size_t size) {
void *x;
x = aom_memalign(DEFAULT_ALIGNMENT, num * size);
if (x) memset(x, 0, num * size);
return x;
}
struct aom_noise_tx_t *aom_noise_tx_malloc(int block_size) {
struct aom_noise_tx_t *noise_tx =
(struct aom_noise_tx_t *)aom_malloc(sizeof(struct aom_noise_tx_t));
if (!noise_tx) return NULL;
memset(noise_tx, 0, sizeof(*noise_tx));
switch (block_size) {
case 2:
noise_tx->fft = aom_fft2x2_float;
noise_tx->ifft = aom_ifft2x2_float;
break;
case 4:
noise_tx->fft = aom_fft4x4_float;
noise_tx->ifft = aom_ifft4x4_float;
break;
case 8:
noise_tx->fft = aom_fft8x8_float;
noise_tx->ifft = aom_ifft8x8_float;
break;
case 16:
noise_tx->fft = aom_fft16x16_float;
noise_tx->ifft = aom_ifft16x16_float;
break;
case 32:
noise_tx->fft = aom_fft32x32_float;
noise_tx->ifft = aom_ifft32x32_float;
break;
default:
aom_free(noise_tx);
fprintf(stderr, "Unsupported block size %d\n", block_size);
return NULL;
}
noise_tx->block_size = block_size;
noise_tx->tx_block = (float *)aom_memalign(
32, 2 * sizeof(*noise_tx->tx_block) * block_size * block_size);
noise_tx->temp = (float *)aom_memalign(
32, 2 * sizeof(*noise_tx->temp) * block_size * block_size);
if (!noise_tx->tx_block || !noise_tx->temp) {
aom_noise_tx_free(noise_tx);
return NULL;
}
return noise_tx;
}
void *aom_malloc(size_t size) { return aom_memalign(DEFAULT_ALIGNMENT, size); }
void av1_encode_tiles_mt(AV1_COMP *cpi) {
AV1_COMMON *const cm = &cpi->common;
const int tile_cols = cm->tile_cols;
const AVxWorkerInterface *const winterface = aom_get_worker_interface();
const int num_workers = AOMMIN(cpi->oxcf.max_threads, tile_cols);
int i;
av1_init_tile_data(cpi);
// Only run once to create threads and allocate thread data.
if (cpi->num_workers == 0) {
CHECK_MEM_ERROR(cm, cpi->workers,
aom_malloc(num_workers * sizeof(*cpi->workers)));
CHECK_MEM_ERROR(cm, cpi->tile_thr_data,
aom_calloc(num_workers, sizeof(*cpi->tile_thr_data)));
for (i = 0; i < num_workers; i++) {
AVxWorker *const worker = &cpi->workers[i];
EncWorkerData *const thread_data = &cpi->tile_thr_data[i];
++cpi->num_workers;
winterface->init(worker);
thread_data->cpi = cpi;
if (i < num_workers - 1) {
// Allocate thread data.
CHECK_MEM_ERROR(cm, thread_data->td,
aom_memalign(32, sizeof(*thread_data->td)));
av1_zero(*thread_data->td);
// Set up pc_tree.
thread_data->td->leaf_tree = NULL;
thread_data->td->pc_tree = NULL;
av1_setup_pc_tree(cm, thread_data->td);
// Set up variance tree if needed.
if (cpi->sf.partition_search_type == VAR_BASED_PARTITION)
av1_setup_var_tree(cm, thread_data->td);
// Allocate frame counters in thread data.
CHECK_MEM_ERROR(cm, thread_data->td->counts,
aom_calloc(1, sizeof(*thread_data->td->counts)));
// Create threads
if (!winterface->reset(worker))
aom_internal_error(&cm->error, AOM_CODEC_ERROR,
"Tile encoder thread creation failed");
} else {
// Main thread acts as a worker and uses the thread data in cpi.
thread_data->td = &cpi->td;
}
winterface->sync(worker);
}
}
for (i = 0; i < num_workers; i++) {
AVxWorker *const worker = &cpi->workers[i];
EncWorkerData *thread_data;
worker->hook = (AVxWorkerHook)enc_worker_hook;
worker->data1 = &cpi->tile_thr_data[i];
worker->data2 = NULL;
thread_data = (EncWorkerData *)worker->data1;
// Before encoding a frame, copy the thread data from cpi.
if (thread_data->td != &cpi->td) {
thread_data->td->mb = cpi->td.mb;
thread_data->td->rd_counts = cpi->td.rd_counts;
}
if (thread_data->td->counts != &cpi->common.counts) {
memcpy(thread_data->td->counts, &cpi->common.counts,
sizeof(cpi->common.counts));
}
#if CONFIG_PALETTE
// Allocate buffers used by palette coding mode.
if (cpi->common.allow_screen_content_tools && i < num_workers - 1) {
MACROBLOCK *x = &thread_data->td->mb;
CHECK_MEM_ERROR(cm, x->palette_buffer,
aom_memalign(16, sizeof(*x->palette_buffer)));
}
#endif // CONFIG_PALETTE
}
// Encode a frame
for (i = 0; i < num_workers; i++) {
AVxWorker *const worker = &cpi->workers[i];
EncWorkerData *const thread_data = (EncWorkerData *)worker->data1;
// Set the starting tile for each thread.
thread_data->start = i;
if (i == cpi->num_workers - 1)
winterface->execute(worker);
else
winterface->launch(worker);
}
// Encoding ends.
for (i = 0; i < num_workers; i++) {
AVxWorker *const worker = &cpi->workers[i];
winterface->sync(worker);
}
for (i = 0; i < num_workers; i++) {
AVxWorker *const worker = &cpi->workers[i];
EncWorkerData *const thread_data = (EncWorkerData *)worker->data1;
// Accumulate counters.
if (i < cpi->num_workers - 1) {
av1_accumulate_frame_counts(&cm->counts, thread_data->td->counts);
accumulate_rd_opt(&cpi->td, thread_data->td);
}
}
}
AV1Decoder *av1_decoder_create(BufferPool *const pool) {
AV1Decoder *volatile const pbi = aom_memalign(32, sizeof(*pbi));
AV1_COMMON *volatile const cm = pbi ? &pbi->common : NULL;
if (!cm) return NULL;
av1_zero(*pbi);
if (setjmp(cm->error.jmp)) {
cm->error.setjmp = 0;
av1_decoder_remove(pbi);
return NULL;
}
cm->error.setjmp = 1;
CHECK_MEM_ERROR(cm, cm->fc,
(FRAME_CONTEXT *)aom_memalign(32, sizeof(*cm->fc)));
CHECK_MEM_ERROR(cm, cm->frame_contexts,
(FRAME_CONTEXT *)aom_memalign(
32, FRAME_CONTEXTS * sizeof(*cm->frame_contexts)));
memset(cm->fc, 0, sizeof(*cm->fc));
memset(cm->frame_contexts, 0, FRAME_CONTEXTS * sizeof(*cm->frame_contexts));
pbi->need_resync = 1;
once(initialize_dec);
// Initialize the references to not point to any frame buffers.
memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
memset(&cm->next_ref_frame_map, -1, sizeof(cm->next_ref_frame_map));
cm->current_video_frame = 0;
pbi->ready_for_new_data = 1;
pbi->common.buffer_pool = pool;
cm->bit_depth = AOM_BITS_8;
cm->dequant_bit_depth = AOM_BITS_8;
cm->alloc_mi = av1_dec_alloc_mi;
cm->free_mi = av1_dec_free_mi;
cm->setup_mi = av1_dec_setup_mi;
av1_loop_filter_init(cm);
#if CONFIG_NCOBMC_ADAPT_WEIGHT
get_default_ncobmc_kernels(cm);
#endif // CONFIG_NCOBMC_ADAPT_WEIGHT
#if CONFIG_AOM_QM
aom_qm_init(cm);
#endif
#if CONFIG_LOOP_RESTORATION
av1_loop_restoration_precal();
#endif // CONFIG_LOOP_RESTORATION
#if CONFIG_ACCOUNTING
pbi->acct_enabled = 1;
aom_accounting_init(&pbi->accounting);
#endif
cm->error.setjmp = 0;
aom_get_worker_interface()->init(&pbi->lf_worker);
return pbi;
}
int aom_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height,
int ss_x, int ss_y,
#if CONFIG_HIGHBITDEPTH
int use_highbitdepth,
#endif
int border, int byte_alignment,
aom_codec_frame_buffer_t *fb,
aom_get_frame_buffer_cb_fn_t cb, void *cb_priv) {
if (ybf) {
const int aom_byte_align = (byte_alignment == 0) ? 1 : byte_alignment;
const int aligned_width = (width + 7) & ~7;
const int aligned_height = (height + 7) & ~7;
const int y_stride = ((aligned_width + 2 * border) + 31) & ~31;
const uint64_t yplane_size =
(aligned_height + 2 * border) * (uint64_t)y_stride + byte_alignment;
const int uv_width = aligned_width >> ss_x;
const int uv_height = aligned_height >> ss_y;
const int uv_stride = y_stride >> ss_x;
const int uv_border_w = border >> ss_x;
const int uv_border_h = border >> ss_y;
const uint64_t uvplane_size =
(uv_height + 2 * uv_border_h) * (uint64_t)uv_stride + byte_alignment;
#if CONFIG_HIGHBITDEPTH
const uint64_t frame_size =
(1 + use_highbitdepth) * (yplane_size + 2 * uvplane_size);
#else
const uint64_t frame_size = yplane_size + 2 * uvplane_size;
#endif // CONFIG_HIGHBITDEPTH
uint8_t *buf = NULL;
if (cb != NULL) {
const int align_addr_extra_size = 31;
const uint64_t external_frame_size = frame_size + align_addr_extra_size;
assert(fb != NULL);
if (external_frame_size != (size_t)external_frame_size) return -1;
// Allocation to hold larger frame, or first allocation.
if (cb(cb_priv, (size_t)external_frame_size, fb) < 0) return -1;
if (fb->data == NULL || fb->size < external_frame_size) return -1;
ybf->buffer_alloc = (uint8_t *)yv12_align_addr(fb->data, 32);
#if defined(__has_feature)
#if __has_feature(memory_sanitizer)
// This memset is needed for fixing the issue of using uninitialized
// value in msan test. It will cause a perf loss, so only do this for
// msan test.
memset(ybf->buffer_alloc, 0, (int)frame_size);
#endif
#endif
} else if (frame_size > (size_t)ybf->buffer_alloc_sz) {
// Allocation to hold larger frame, or first allocation.
aom_free(ybf->buffer_alloc);
ybf->buffer_alloc = NULL;
if (frame_size != (size_t)frame_size) return -1;
ybf->buffer_alloc = (uint8_t *)aom_memalign(32, (size_t)frame_size);
if (!ybf->buffer_alloc) return -1;
ybf->buffer_alloc_sz = (size_t)frame_size;
// This memset is needed for fixing valgrind error from C loop filter
// due to access uninitialized memory in frame border. It could be
// removed if border is totally removed.
memset(ybf->buffer_alloc, 0, ybf->buffer_alloc_sz);
}
/* Only support allocating buffers that have a border that's a multiple
* of 32. The border restriction is required to get 16-byte alignment of
* the start of the chroma rows without introducing an arbitrary gap
* between planes, which would break the semantics of things like
* aom_img_set_rect(). */
if (border & 0x1f) return -3;
ybf->y_crop_width = width;
ybf->y_crop_height = height;
ybf->y_width = aligned_width;
ybf->y_height = aligned_height;
ybf->y_stride = y_stride;
ybf->uv_crop_width = (width + ss_x) >> ss_x;
ybf->uv_crop_height = (height + ss_y) >> ss_y;
ybf->uv_width = uv_width;
ybf->uv_height = uv_height;
ybf->uv_stride = uv_stride;
ybf->border = border;
ybf->frame_size = (size_t)frame_size;
ybf->subsampling_x = ss_x;
ybf->subsampling_y = ss_y;
buf = ybf->buffer_alloc;
#if CONFIG_HIGHBITDEPTH
if (use_highbitdepth) {
// Store uint16 addresses when using 16bit framebuffers
buf = CONVERT_TO_BYTEPTR(ybf->buffer_alloc);
ybf->flags = YV12_FLAG_HIGHBITDEPTH;
} else {
ybf->flags = 0;
}
#endif // CONFIG_HIGHBITDEPTH
ybf->y_buffer = (uint8_t *)yv12_align_addr(
buf + (border * y_stride) + border, aom_byte_align);
ybf->u_buffer = (uint8_t *)yv12_align_addr(
buf + yplane_size + (uv_border_h * uv_stride) + uv_border_w,
aom_byte_align);
ybf->v_buffer =
(uint8_t *)yv12_align_addr(buf + yplane_size + uvplane_size +
(uv_border_h * uv_stride) + uv_border_w,
aom_byte_align);
#if CONFIG_HIGHBITDEPTH && CONFIG_GLOBAL_MOTION
if (ybf->y_buffer_8bit) {
free(ybf->y_buffer_8bit);
ybf->y_buffer_8bit = NULL;
}
#endif
ybf->corrupted = 0; /* assume not corrupted by errors */
return 0;
}
return -2;
}