jessikitty/unity
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Remove duplicate endpoints and selectors from the codebooks

Browse Source 
This change significantly improves the compression ratio.

Explanation:
By default, the size of the endpoint and selector codebooks is calculated based on the number of blocks in the image and the quality parameter, while the actual complexity of the image does not affect the initial codebook size. So the target codebook size is selected in such a way, that even complex images can be approximated well enough. At the same time, normally, the lower is the complexity of the image, the higher is the density of the quantized vectors. Considering that vector quantization is performed using floating point computations, and the quantized endpoints have integer components, high density of quantized vectors will result in large number of duplicate endpoints. As the result, some identical endpoints are being represented by multiple different indices, which significantly affects the compression ratio. Note that this is not the case for selectors, as their corresponding vector components are rounded after quantization, but instead it leads to some duplicate selectors in the codebook being not used. In the modified version of the algorithm all the duplicate codebook entries are merged together, unused entries are removed from the codebooks, the endpoint and selector indices are updated accordingly.

Testing:
The modified algorithm has been tested on the Kodak test set using 64-bit build with default settings (running on Windows 10, i7-4790, 3.6GHz). All the decompressed test images are identical to the images being compressed and decompressed using original version of Crunch.

[Compressing Kodak set without mipmaps]
Original: 1582222 bytes / 28.835 sec
Modified: 1494630 bytes / 25.637 sec
Improvement: 5.54% (compression ratio) / 11.09% (compression time)

[Compressing Kodak set with mipmaps]
Original: 2065243 bytes / 36.875 sec
Modified: 1946533 bytes / 33.546 sec
Improvement: 5.75% (compression ratio) / 9.03% (compression time)
This commit is contained in:
Alexander Suvorov
2017-05-05 20:06:00 +02:00
parent ef540e54de
commit 5258727545
2 changed files with 146 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
bin/crunch_x64.exe
BIN
View File
Binary file not shown.
crnlib/crn_dxt_hc.cpp
+146
View File
@@ -1729,6 +1729,37 @@ bool dxt_hc::refine_quantized_color_selectors() {
console::info("Total refined pixels: %u, selectors: %u out of %u", total_refined_pixels, total_refined_selectors, total_selectors);
#endif
uint selector_count = 0;
hash_map<uint32, uint> packed_selectors;
for (uint i = 0; i < m_color_selectors.size(); i++) {
if (m_chunk_blocks_using_color_selectors[i].size()) {
uint32 packed_selector = 0;
for (uint s = 0; s < 16; s++)
packed_selector |= m_color_selectors[i].get_by_index(s) << (s << 1);
hash_map<uint32, uint>::insert_result insert_result = packed_selectors.insert(packed_selector, selector_count);
if (insert_result.second) {
if (selector_count != i) {
m_color_selectors[selector_count] = m_color_selectors[i];
m_chunk_blocks_using_color_selectors[selector_count].swap(m_chunk_blocks_using_color_selectors[i]);
for (uint b = 0; b < m_chunk_blocks_using_color_selectors[selector_count].size(); b++) {
const block_id& id = m_chunk_blocks_using_color_selectors[selector_count][b];
m_compressed_chunks[cColorChunks][id.m_chunk_index].m_selector_cluster_index[id.m_block_y][id.m_block_x] = selector_count;
}
}
selector_count++;
} else {
for (uint b = 0; b < m_chunk_blocks_using_color_selectors[i].size(); b++) {
const block_id& id = m_chunk_blocks_using_color_selectors[i][b];
m_compressed_chunks[cColorChunks][id.m_chunk_index].m_selector_cluster_index[id.m_block_y][id.m_block_x] = insert_result.first->second;
}
m_chunk_blocks_using_color_selectors[insert_result.first->second].append(m_chunk_blocks_using_color_selectors[i]);
m_chunk_blocks_using_color_selectors[i].clear();
}
}
}
m_color_selectors.resize(selector_count);
m_chunk_blocks_using_color_selectors.resize(selector_count);
return true;
}
@@ -1824,6 +1855,37 @@ bool dxt_hc::refine_quantized_alpha_selectors() {
console::info("Total refined pixels: %u, selectors: %u out of %u", total_refined_pixels, total_refined_selectors, total_selectors);
#endif
uint selector_count = 0;
hash_map<uint64, uint> packed_selectors;
for (uint i = 0; i < m_alpha_selectors.size(); i++) {
if (m_chunk_blocks_using_alpha_selectors[i].size()) {
uint64 packed_selector = 0;
for (uint s = 0; s < 16; s++)
packed_selector |= (uint64)m_alpha_selectors[i].get_by_index(s) << (s << 2);
hash_map<uint64, uint>::insert_result insert_result = packed_selectors.insert(packed_selector, selector_count);
if (insert_result.second) {
if (selector_count != i) {
m_alpha_selectors[selector_count] = m_alpha_selectors[i];
m_chunk_blocks_using_alpha_selectors[selector_count].swap(m_chunk_blocks_using_alpha_selectors[i]);
for (uint b = 0; b < m_chunk_blocks_using_alpha_selectors[selector_count].size(); b++) {
const block_id& id = m_chunk_blocks_using_alpha_selectors[selector_count][b];
m_compressed_chunks[cAlpha0Chunks + id.m_alpha_index][id.m_chunk_index].m_selector_cluster_index[id.m_block_y][id.m_block_x] = selector_count;
}
}
selector_count++;
} else {
for (uint b = 0; b < m_chunk_blocks_using_alpha_selectors[i].size(); b++) {
const block_id& id = m_chunk_blocks_using_alpha_selectors[i][b];
m_compressed_chunks[cAlpha0Chunks + id.m_alpha_index][id.m_chunk_index].m_selector_cluster_index[id.m_block_y][id.m_block_x] = insert_result.first->second;
}
m_chunk_blocks_using_alpha_selectors[insert_result.first->second].append(m_chunk_blocks_using_alpha_selectors[i]);
m_chunk_blocks_using_alpha_selectors[i].clear();
}
}
}
m_alpha_selectors.resize(selector_count);
m_chunk_blocks_using_alpha_selectors.resize(selector_count);
return true;
}
@@ -1933,6 +1995,47 @@ bool dxt_hc::refine_quantized_color_endpoints() {
console::info("Total refined pixels: %u, endpoints: %u out of %u", total_refined_pixels, total_refined_tiles, m_color_clusters.size());
#endif
uint cluster_count = 0;
hash_map<uint64, uint> packed_clusters;
for (uint i = 0; i < m_color_clusters.size(); i++) {
tile_cluster& cluster = m_color_clusters[i];
if (cluster.m_tiles.size()) {
uint64 packed_cluster = cluster.m_first_endpoint | cluster.m_second_endpoint << 16 | (cluster.m_alpha_encoding ? 1ULL << 32 : 0);
hash_map<uint64, uint>::insert_result insert_result = packed_clusters.insert(packed_cluster, cluster_count);
if (insert_result.second) {
if (cluster_count != i) {
tile_cluster& destination_cluster = m_color_clusters[cluster_count];
destination_cluster.m_alpha_encoding = cluster.m_alpha_encoding;
destination_cluster.m_error = cluster.m_error;
destination_cluster.m_first_endpoint = cluster.m_first_endpoint;
destination_cluster.m_second_endpoint = cluster.m_second_endpoint;
destination_cluster.m_tiles.swap(cluster.m_tiles);
for (uint t = 0; t < destination_cluster.m_tiles.size(); t++) {
const uint chunk_index = destination_cluster.m_tiles[t].first;
const uint tile_index = destination_cluster.m_tiles[t].second;
compressed_tile& tile = m_compressed_chunks[cColorChunks][chunk_index].m_quantized_tiles[tile_index];
tile.m_first_endpoint = destination_cluster.m_first_endpoint;
tile.m_second_endpoint = destination_cluster.m_second_endpoint;
tile.m_endpoint_cluster_index = cluster_count;
}
}
cluster_count++;
} else {
tile_cluster& destination_cluster = m_color_clusters[insert_result.first->second];
for (uint t = 0; t < cluster.m_tiles.size(); t++) {
const uint chunk_index = cluster.m_tiles[t].first;
const uint tile_index = cluster.m_tiles[t].second;
compressed_tile& tile = m_compressed_chunks[cColorChunks][chunk_index].m_quantized_tiles[tile_index];
tile.m_endpoint_cluster_index = insert_result.first->second;
}
destination_cluster.m_error += cluster.m_error;
destination_cluster.m_tiles.append(cluster.m_tiles);
cluster.m_tiles.clear();
}
}
}
m_color_clusters.resize(cluster_count);
return true;
}
@@ -2044,6 +2147,49 @@ bool dxt_hc::refine_quantized_alpha_endpoints() {
console::info("Total refined pixels: %u, endpoints: %u out of %u", total_refined_pixels, total_refined_tiles, m_alpha_clusters.size());
#endif
uint cluster_count = 0;
hash_map<uint64, uint> packed_clusters;
for (uint i = 0; i < m_alpha_clusters.size(); i++) {
tile_cluster& cluster = m_alpha_clusters[i];
if (cluster.m_tiles.size()) {
uint64 packed_cluster = cluster.m_first_endpoint | cluster.m_second_endpoint << 16 | (cluster.m_alpha_encoding ? 1ULL << 32 : 0);
hash_map<uint64, uint>::insert_result insert_result = packed_clusters.insert(packed_cluster, cluster_count);
if (insert_result.second) {
if (cluster_count != i) {
tile_cluster& destination_cluster = m_alpha_clusters[cluster_count];
destination_cluster.m_alpha_encoding = cluster.m_alpha_encoding;
destination_cluster.m_error = cluster.m_error;
destination_cluster.m_first_endpoint = cluster.m_first_endpoint;
destination_cluster.m_second_endpoint = cluster.m_second_endpoint;
destination_cluster.m_tiles.swap(cluster.m_tiles);
for (uint t = 0; t < destination_cluster.m_tiles.size(); t++) {
const uint chunk_index = destination_cluster.m_tiles[t].first;
const uint tile_index = destination_cluster.m_tiles[t].second & 0xFFFFU;
const uint alpha_index = destination_cluster.m_tiles[t].second >> 16U;
compressed_tile& tile = m_compressed_chunks[cAlpha0Chunks + alpha_index][chunk_index].m_quantized_tiles[tile_index];
tile.m_first_endpoint = destination_cluster.m_first_endpoint;
tile.m_second_endpoint = destination_cluster.m_second_endpoint;
tile.m_endpoint_cluster_index = cluster_count;
}
}
cluster_count++;
} else {
tile_cluster& destination_cluster = m_alpha_clusters[insert_result.first->second];
for (uint t = 0; t < cluster.m_tiles.size(); t++) {
const uint chunk_index = cluster.m_tiles[t].first;
const uint tile_index = cluster.m_tiles[t].second & 0xFFFFU;
const uint alpha_index = cluster.m_tiles[t].second >> 16U;
compressed_tile& tile = m_compressed_chunks[cAlpha0Chunks + alpha_index][chunk_index].m_quantized_tiles[tile_index];
tile.m_endpoint_cluster_index = insert_result.first->second;
}
destination_cluster.m_error += cluster.m_error;
destination_cluster.m_tiles.append(cluster.m_tiles);
cluster.m_tiles.clear();
}
}
}
m_alpha_clusters.resize(cluster_count);
return true;
}