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 +10 -9
2 changed files with 10 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files
Showing only changes of commit 67b52758bb - Show all commits

4
source/blender/blenlib/BLI_index_mask2.hh
View File

@ -41,7 +41,7 @@ struct RawMaskIterator {
*/
struct Chunk {
int16_t segments_num;
const int16_t **segment_indices;
const int16_t **indices_by_segment;
const int16_t *segment_sizes_cumulative;
RawChunkIterator end_iterator() const;
@ -476,7 +476,7 @@ template<typename Fn> inline void IndexMask::foreach_raw_segment(Fn &&fn) const
const int16_t stored_segment_size = cumulative_segment_size - prev_cumulative_segment_size;
const bool is_last_segment = is_last_chunk & (segment_i == final_segment_i);
const int16_t segment_drop_back = is_last_segment * final_drop_back;
const int16_t *indices_in_segment = chunk.segment_indices[segment_i] + segment_drop_front;
const int16_t *indices_in_segment = chunk.indices_by_segment[segment_i] + segment_drop_front;
const int16_t segment_size = stored_segment_size - segment_drop_front - segment_drop_back;
const Span<int16_t> indices_span{indices_in_segment, segment_size};

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

@ -39,7 +39,7 @@ const IndexMask &get_static_index_mask_for_min_size(const int64_t min_size)
for (const int64_t i : range) {
Chunk &chunk = chunks_array[i];
chunk.segments_num = 1;
chunk.segment_indices = &static_offsets;
chunk.indices_by_segment = &static_offsets;
chunk.segment_sizes_cumulative = static_segment_sizes_cumulative;
chunk_ids_array[i] = i;
@ -227,7 +227,7 @@ template<typename T> IndexMask to_index_mask(const Span<T> indices, ResourceScop
LinearAllocator<> &allocator = allocator_by_thread.local();
MutableSpan<int16_t> remaining_indices = allocator.allocate_array<int16_t>(
index_allocations_num);
MutableSpan<const int16_t *> remaining_segment_indices_pointers =
MutableSpan<const int16_t *> remaining_indices_by_segment =
allocator.allocate_array<const int16_t *>(segments_in_chunks.size());
MutableSpan<int16_t> remaining_segment_sizes_cumulative = allocator.allocate_array<int16_t>(
segments_in_chunks.size() + slice.size());
@ -240,8 +240,8 @@ template<typename T> IndexMask to_index_mask(const Span<T> indices, ResourceScop
Chunk &chunk = chunks[slice[i]];
const int64_t chunk_offset = chunk_ids[slice[i]] * chunk_capacity;
MutableSpan<const int16_t *> segment_indices = take_front_and_drop(
remaining_segment_indices_pointers, segments_num);
MutableSpan<const int16_t *> indices_by_segment = take_front_and_drop(
remaining_indices_by_segment, segments_num);
MutableSpan<int16_t> segment_sizes_cumulative = take_front_and_drop(
remaining_segment_sizes_cumulative, segments_num + 1);
@ -252,7 +252,8 @@ template<typename T> IndexMask to_index_mask(const Span<T> indices, ResourceScop
if (std::holds_alternative<IndexRange>(segment)) {
const IndexRange range_in_segment = std::get<IndexRange>(segment);
segment_indices[segment_i] = static_offsets + (range_in_segment.first() - chunk_offset);
indices_by_segment[segment_i] = static_offsets +
(range_in_segment.first() - chunk_offset);
cumulative_size += range_in_segment.size();
}
else {
@ -263,13 +264,13 @@ template<typename T> IndexMask to_index_mask(const Span<T> indices, ResourceScop
new_indices[index_in_segment] = int16_t(indices_in_segment[index_in_segment] -
chunk_offset);
}
segment_indices[segment_i] = new_indices.data();
indices_by_segment[segment_i] = new_indices.data();
cumulative_size += indices_in_segment.size();
}
}
segment_sizes_cumulative[segments_num] = int16_t(cumulative_size);
chunk.segment_indices = segment_indices.data();
chunk.indices_by_segment = indices_by_segment.data();
chunk.segment_sizes_cumulative = segment_sizes_cumulative.data();
}
});