blender
/
blender
136
407
Fork 589
Code Issues 6.3k Pull Requests 767 Projects 19 Wiki Activity

BLI: refactor IndexMask for better performance and memory usage #104629

Merged
Jacques Lucke merged 254 commits from JacquesLucke/blender:index-mask-refactor into main 2023-05-24 18:11:47 +02:00
Conversation 6 Commits 254 Files Changed 182 +13 -13
2 changed files with 13 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files
Showing only changes of commit f3f69f6e0e - Show all commits

11
source/blender/blenlib/BLI_index_mask2.hh
View File

@ -56,7 +56,7 @@ struct IndexMaskData {
int64_t chunks_num;
int64_t indices_num;
const Chunk *chunks;
const int64_t *chunk_offsets;
const int64_t *chunk_ids;
const int64_t *chunk_sizes_cumulative;
RawChunkIterator begin_it;
RawChunkIterator end_it;
@ -319,7 +319,7 @@ inline IndexMask::IndexMask()
data_.chunks_num = 0;
data_.chunks = nullptr;
data_.chunk_offsets = nullptr;
data_.chunk_ids = nullptr;
data_.chunk_sizes_cumulative = cumulative_sizes_for_empty_mask;
data_.begin_it.segment_i = 0;
data_.begin_it.index_in_segment = 0;
@ -348,7 +348,7 @@ inline IndexMask::IndexMask(const IndexRange range)
data_.chunks_num = last_chunk_id - first_chunk_id + 1;
data_.indices_num = range.size();
data_.chunks -= first_chunk_id;
data_.chunk_offsets -= first_chunk_id;
data_.chunk_ids -= first_chunk_id;
data_.chunk_sizes_cumulative -= first_chunk_id;
data_.begin_it.segment_i = 0;
data_.begin_it.index_in_segment = range.first() & chunk_mask_low;
@ -399,7 +399,7 @@ inline IndexMask IndexMask::slice(const IndexRange range) const
sliced.data_.chunks_num = last_it.chunk_i - first_it.chunk_i + 1;
sliced.data_.indices_num = range.size();
sliced.data_.chunks = data_.chunks + first_it.chunk_i;
sliced.data_.chunk_offsets = data_.chunk_offsets + first_it.chunk_i;
sliced.data_.chunk_ids = data_.chunk_ids + first_it.chunk_i;
sliced.data_.chunk_sizes_cumulative = data_.chunk_sizes_cumulative + first_it.chunk_i;
sliced.data_.begin_it = first_it.chunk_it;
sliced.data_.end_it.segment_i = last_it.chunk_it.segment_i;
@ -465,9 +465,10 @@ template<typename Fn> inline void IndexMask::foreach_raw_segment(Fn &&fn) const
int64_t counter = 0;
while (chunk_i < data_.chunks_num) {
const Chunk &chunk = data_.chunks[chunk_i];
const int64_t chunk_id = data_.chunk_ids[chunk_i];
const bool is_last_chunk = (chunk_i == data_.chunks_num - 1);
const int16_t segments_num = is_last_chunk ? final_segments_num : chunk.segments_num;
const int64_t offset = data_.chunk_offsets[chunk_i];
const int64_t offset = max_chunk_size + chunk_id;
int16_t prev_cumulative_segment_size = chunk.segment_sizes_cumulative[segment_i];
while (segment_i < segments_num) {
const int16_t next_segment_i = segment_i + 1;

15
source/blender/blenlib/intern/index_mask2.cc
View File

@ -29,7 +29,7 @@ const IndexMask &get_static_index_mask_for_min_size(const int64_t min_size)
static IndexMask static_mask = []() {
static Array<Chunk> chunks_array(chunks_num);
static Array<int64_t> chunk_offsets_array(chunks_num);
static Array<int64_t> chunk_ids_array(chunks_num);
static Array<int64_t> chunk_sizes_cumulative(chunks_num + 1);
static const int16_t *static_offsets = get_static_indices_array().data();
@ -42,7 +42,7 @@ const IndexMask &get_static_index_mask_for_min_size(const int64_t min_size)
chunk.segment_indices = &static_offsets;
chunk.segment_sizes_cumulative = static_segment_sizes_cumulative;
chunk_offsets_array[i] = i * max_chunk_size;
chunk_ids_array[i] = i;
chunk_sizes_cumulative[i] = i * max_chunk_size;
}
});
@ -53,7 +53,7 @@ const IndexMask &get_static_index_mask_for_min_size(const int64_t min_size)
mask_data.chunks_num = chunks_num;
mask_data.indices_num = max_size;
mask_data.chunks = chunks_array.data();
mask_data.chunk_offsets = chunk_offsets_array.data();
mask_data.chunk_ids = chunk_ids_array.data();
mask_data.chunk_sizes_cumulative = chunk_sizes_cumulative.data();
mask_data.begin_it.segment_i = 0;
mask_data.begin_it.index_in_segment = 0;
@ -172,7 +172,7 @@ template<typename T> IndexMask to_index_mask(const Span<T> indices, ResourceScop
scope.construct<threading::EnumerableThreadSpecific<LinearAllocator<>>>();
MutableSpan<Chunk> chunks = main_allocator.allocate_array<Chunk>(chunks_num);
MutableSpan<int64_t> chunk_offsets = main_allocator.allocate_array<int64_t>(chunks_num);
MutableSpan<int64_t> chunk_ids = main_allocator.allocate_array<int64_t>(chunks_num);
static const int16_t *static_offsets = get_static_indices_array().data();
@ -194,8 +194,7 @@ template<typename T> IndexMask to_index_mask(const Span<T> indices, ResourceScop
const int64_t chunk_id = index_to_chunk_id(int64_t(indices_in_chunk[0]));
BLI_assert(chunk_id == index_to_chunk_id(int64_t(indices_in_chunk.last())));
Chunk &chunk = chunks[i];
const int64_t chunk_offset = max_chunk_size * chunk_id;
chunk_offsets[i] = chunk_offset;
chunk_ids[i] = chunk_id;
const int16_t segments_in_chunk_num = int16_t(
split_to_ranges_and_spans(indices_in_chunk, 16, segments_in_chunks));
@ -239,7 +238,7 @@ template<typename T> IndexMask to_index_mask(const Span<T> indices, ResourceScop
const IndexRange segments_in_chunk = segments_by_chunk[i];
const int16_t segments_num = int16_t(segments_in_chunk.size());
Chunk &chunk = chunks[slice[i]];
const int64_t chunk_offset = chunk_offsets[slice[i]];
const int64_t chunk_offset = chunk_ids[slice[i]] * max_chunk_size;
MutableSpan<const int16_t *> segment_indices = take_front_and_drop(
remaining_segment_indices_pointers, segments_num);
@ -289,7 +288,7 @@ template<typename T> IndexMask to_index_mask(const Span<T> indices, ResourceScop
mask_data.chunks_num = chunks_num;
mask_data.indices_num = indices.size();
mask_data.chunks = chunks.data();
mask_data.chunk_offsets = chunk_offsets.data();
mask_data.chunk_ids = chunk_ids.data();
mask_data.chunk_sizes_cumulative = chunk_sizes_cumulative.data();
mask_data.begin_it = {0, 0};
mask_data.end_it = chunks.last().end_iterator();