rust/tantivy
1
0
Fork 0
mirror of https://github.com/quickwit-oss/tantivy.git synced 2025-12-30 14:02:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

Compare commits

base: rust:0.2.0
Branches Tags
rust:main rust:bucket_id_agg rust:stuhood.upstream-erased-sort-computer rust:trinity.pointard/ff-query rust:paradedb/fix-agg-validation rust:flatheadmill-geo rust:paradedb-paradedb/fix-overflow-issue rust:low_card_optimization rust:paul.masurel/clippy rust:qw-airmail-20250916 rust:unit-test-aggregation-sum-other rust:clippy-and-cleanups rust:trinity.pointard/opti-sstable-sorted-ords-to-term rust:release_0.24 rust:dependabot/cargo/rand-0.9.1 rust:dependabot/cargo/lru-0.14.0 rust:dependabot/cargo/fs4-0.13.1 rust:tagged-user-input-ast rust:1686a/parse-agg-res rust:dependabot/cargo/rand_distr-0.5.1 rust:issue/2411-adding-panic-handler-to-rayon-merge-thread-pool rust:dependabot/cargo/nom-8 rust:upgrade_deps rust:test_parser rust:dependabot/github_actions/codecov/codecov-action-5 rust:binggan-0.12.0 rust:agg_format rust:path_collector rust:test_order_bug rust:test_columnar_multi rust:columnar_optionalmask_in_multivalues rust:prefix-phrase-query-optim rust:tantivy-with-oneshot rust:update_example rust:store-reader-enumerate rust:document-deserialize-seed rust:check_doc_mem rust:experimental-do-not-merge rust:expdonotmerge rust:cherry-pick-vec-instead-of-btreemap rust:validate_length rust:raphael_bench_avx_filter rust:raphael_optim_memset2 rust:query-field-name-avoid-alloc rust:check_bug rust:missing_test rust:list_fields rust:block-cache-control rust:trinity--multilayer-sstable rust:tracing-tantivy2 rust:tracing-tantivy rust:index_writer_err rust:default_fast_field_tokenizer rust:fmassot/add-preserve-original-ascii-folding rust:numeric_with_str_col rust:straightforward-simplify rust:fmassot/fake-test-branch rust:fmassot/dynamic-tokenizer-poc-follow-up rust:fmassot/add-box-token-filter-and-refactor rust:dynamic-tokenizer-proof-of-concept rust:fmassot/add-box-tokenizer-and-token-filter rust:tokenizer-api-0.1.1 rust:addconversion rust:removed-extra-enum-alternative-for-datetime-precision rust:issue/1981-reproduce-bm25-bug rust:tokenizer_docs rust:merge-proptests rust:columnar-todo rust:quickwit-0.5 rust:remove-byteorder rust:remove_dyn rust:pr/alexcole/1855-inject-bm25-statistics rust:evance-expose-build-query-from-user-input-ast rust:missing-specialized-method-arcdyncolumn rust:fast-u64-range-query rust:0.19.2 rust:columnar-cli2 rust:trinity--pub-term-as-slice rust:typed-column3 rust:columnar-merge3 rust:columnar-merge-exploring-fn rust:use_columnar rust:typed-column2 rust:term_agg_bucket_limit rust:range rust:typed-column rust:columnar-main rust:forced-types rust:u128-columnar rust:use_stats rust:truncation_comment rust:0.19.1 rust:trinity/yoke-documents rust:column-handle rust:columnar-cli rust:columnar rust:revert-1711-sparse_dense_index rust:commit-changes rust:sparse_dense_index_headroom rust:optional_column rust:no-column-for-multivalued rust:fmassot--bench-hdfs-with-array rust:bugfix-position-broken rust:float rust:simplify rust:debug_time rust:sparse_codecs rust:fix_estimate rust:column-reader rust:column-reader-b rust:columnreader-simple rust:col-trait-refact rust:ip_fastfield rust:column-trait-2 rust:refact-dyn-column rust:column-trait rust:refact-codec-trait rust:pr/PSeitz/1485-1 rust:fastfieldcodec rust:termmap_per_field rust:fastfield-codecs-tests rust:rayon-indexing-writers rust:issue/1251 rust:index-bench rust:fix_open_bytes_index rust:warming rust:revert-1222-issue/1195 rust:add-random-bench-for-fast-field rust:wasm-friendly-exploratory rust:revert-1159-patch-1 rust:debug-position-panic-0.15.3 rust:fix-issue-1111 rust:remove_rand rust:unused-deps rust:hfb/phrase_match_slop rust:0.15.3 rust:expose-min-max-on-multi-fast-field-reader rust:revert-1094-expose-min-max-on-multi-fast-field-reader rust:hotfix-1088 rust:owned-bytes-instead-of-rawdoc rust:rihardsk-date-ranges rust:issue/997 rust:0.13.3 rust:fail-fast-store-checkpoint rust:debugging-store rust:issue/reproduce-897 rust:issue/969-reproduce rust:troublescooter-static rust:intermediate-mergeable-troublescooter rust:barrotsteindev-filter-collector-tpredicatevalue rust:issue/938b rust:issue/938 rust:barrotsteindev-filter-collector rust:issue/922 rust:issue/896 rust:quickfix-explain rust:slog rust:limit-random-seeks rust:0.13.1 rust:issue/866b rust:bugfix-unittest-macos rust:githubactions rust:fieldnorm_readers rust:blockwand rust:nodeffeatfails rust:removedaliveiterator rust:robyoung-introduce-offset-to-top-docs rust:specialization-of-foreach-for-union-scorer rust:elshize-bmw rust:cutt rust:nrt rust:storewriter-out rust:segment_writer_in_ram2 rust:polling rust:audunhalland-query-boost rust:poll-meta rust:segment_writer_in_ram rust:bundle rust:petr-tik-n662_footer_is_incompatible_impl rust:race-condition-in-tests rust:hotfix-0.10.3-issue681 rust:remove-tru rust:kkoziara-pre-tokenized-text rust:issue/681 rust:issue/680 rust:refactoring/field rust:updating/uuid rust:fix-bench rust:hotfix-656 rust:readd-macos-ci rust:incremental-search rust:crate-division rust:streamer-with-state rust:criterion rust:python-binding-changes rust:bump-version rust:gh-pages rust:issue/554 rust:perf/experimental-union-for_each rust:issue/526b rust:softcommits rust:0.9.1-hotfix rust:revert-521-fix-non-english-stemmers rust:issue/500-hotfix rust:hotfix/0.8.2 rust:issue/483 rust:bug/merge-deted rust:issue/407 rust:issue/457 rust:segment_fruits rust:doc rust:dds/lenient rust:issue/webasm rust:issue/396 rust:broken_collector_with_result rust:dss/automaton-weight-suggestion rust:wasm rust:common-crawl rust:tantivy-imhotep
rust:0.25.0 rust:0.22.1 rust:0.24.2 rust:0.24.1 rust:tantivy-query-grammar-v0.24.0 rust:tantivy-columnar-v0.5.0 rust:tantivy-bitpacker-v0.8.0 rust:ownedbytes-v0.9.0 rust:tantivy-stacker-v0.5.0 rust:tantivy-sstable-v0.5.0 rust:tantivy-common-v0.9.0 rust:tantivy-tokenizer-api-v0.5.0 rust:0.24 rust:qw-airmail-20250224-hotfix-merge-panic rust:qw-airmail-20250219-hotfix-merge-panic rust:qw-airmail-202406 rust:qw-airmail-20240611 rust:qw-airmail-20240606 rust:quickwit-rev rust:0.22.0 rust:tantivy-stacker-v0.3.0 rust:ownedbytes-v0.7.0 rust:tantivy-tokenizer-api-v0.3.0 rust:tantivy-columnar-v0.3.0 rust:tantivy-common-v0.7.0 rust:tantivy-query-grammar-v0.22.0 rust:tantivy-sstable-v0.3.0 rust:tantivy-bitpacker-v0.6.0 rust:0.21.1 rust:0.21 rust:0.20.2 rust:0.20.1 rust:0.20 rust:quickwit-0.5-rev rust:0.19.2 rust:0.19.1 rust:0.19 rust:0.18.1 rust:0.18 rust:0.17 rust:0.16.1 rust:0.15.3 rust:0.15.2 rust:0.15.1 rust:0.15 rust:0.14 rust:0.13.3 rust:0.13.2 rust:0.13.1 rust:0.13 rust:0.12 rust:0.11.3 rust:0.11.1 rust:0.11 rust:0.10.3 rust:0.1 rust:0.10.2 rust:0.10.1 rust:0.10.0 rust:0.9.1 rust:0.9 rust:0.8.2 rust:0.8.2b rust:0.8.1 rust:0.8.0 rust:0.7.2 rust:0.7.1 rust:0.7.0 rust:0.6.1 rust:0.6.0 rust:0.5.2 rust:0.5.1 rust:0.5.0 rust:0.4.3 rust:0.4.2 rust:0.4.0 rust:0.3.1 rust:0.3.0 rust:0.2.0
...
compare: rust:0.4.3
Branches Tags
rust:main rust:bucket_id_agg rust:stuhood.upstream-erased-sort-computer rust:trinity.pointard/ff-query rust:paradedb/fix-agg-validation rust:flatheadmill-geo rust:paradedb-paradedb/fix-overflow-issue rust:low_card_optimization rust:paul.masurel/clippy rust:qw-airmail-20250916 rust:unit-test-aggregation-sum-other rust:clippy-and-cleanups rust:trinity.pointard/opti-sstable-sorted-ords-to-term rust:release_0.24 rust:dependabot/cargo/rand-0.9.1 rust:dependabot/cargo/lru-0.14.0 rust:dependabot/cargo/fs4-0.13.1 rust:tagged-user-input-ast rust:1686a/parse-agg-res rust:dependabot/cargo/rand_distr-0.5.1 rust:issue/2411-adding-panic-handler-to-rayon-merge-thread-pool rust:dependabot/cargo/nom-8 rust:upgrade_deps rust:test_parser rust:dependabot/github_actions/codecov/codecov-action-5 rust:binggan-0.12.0 rust:agg_format rust:path_collector rust:test_order_bug rust:test_columnar_multi rust:columnar_optionalmask_in_multivalues rust:prefix-phrase-query-optim rust:tantivy-with-oneshot rust:update_example rust:store-reader-enumerate rust:document-deserialize-seed rust:check_doc_mem rust:experimental-do-not-merge rust:expdonotmerge rust:cherry-pick-vec-instead-of-btreemap rust:validate_length rust:raphael_bench_avx_filter rust:raphael_optim_memset2 rust:query-field-name-avoid-alloc rust:check_bug rust:missing_test rust:list_fields rust:block-cache-control rust:trinity--multilayer-sstable rust:tracing-tantivy2 rust:tracing-tantivy rust:index_writer_err rust:default_fast_field_tokenizer rust:fmassot/add-preserve-original-ascii-folding rust:numeric_with_str_col rust:straightforward-simplify rust:fmassot/fake-test-branch rust:fmassot/dynamic-tokenizer-poc-follow-up rust:fmassot/add-box-token-filter-and-refactor rust:dynamic-tokenizer-proof-of-concept rust:fmassot/add-box-tokenizer-and-token-filter rust:tokenizer-api-0.1.1 rust:addconversion rust:removed-extra-enum-alternative-for-datetime-precision rust:issue/1981-reproduce-bm25-bug rust:tokenizer_docs rust:merge-proptests rust:columnar-todo rust:quickwit-0.5 rust:remove-byteorder rust:remove_dyn rust:pr/alexcole/1855-inject-bm25-statistics rust:evance-expose-build-query-from-user-input-ast rust:missing-specialized-method-arcdyncolumn rust:fast-u64-range-query rust:0.19.2 rust:columnar-cli2 rust:trinity--pub-term-as-slice rust:typed-column3 rust:columnar-merge3 rust:columnar-merge-exploring-fn rust:use_columnar rust:typed-column2 rust:term_agg_bucket_limit rust:range rust:typed-column rust:columnar-main rust:forced-types rust:u128-columnar rust:use_stats rust:truncation_comment rust:0.19.1 rust:trinity/yoke-documents rust:column-handle rust:columnar-cli rust:columnar rust:revert-1711-sparse_dense_index rust:commit-changes rust:sparse_dense_index_headroom rust:optional_column rust:no-column-for-multivalued rust:fmassot--bench-hdfs-with-array rust:bugfix-position-broken rust:float rust:simplify rust:debug_time rust:sparse_codecs rust:fix_estimate rust:column-reader rust:column-reader-b rust:columnreader-simple rust:col-trait-refact rust:ip_fastfield rust:column-trait-2 rust:refact-dyn-column rust:column-trait rust:refact-codec-trait rust:pr/PSeitz/1485-1 rust:fastfieldcodec rust:termmap_per_field rust:fastfield-codecs-tests rust:rayon-indexing-writers rust:issue/1251 rust:index-bench rust:fix_open_bytes_index rust:warming rust:revert-1222-issue/1195 rust:add-random-bench-for-fast-field rust:wasm-friendly-exploratory rust:revert-1159-patch-1 rust:debug-position-panic-0.15.3 rust:fix-issue-1111 rust:remove_rand rust:unused-deps rust:hfb/phrase_match_slop rust:0.15.3 rust:expose-min-max-on-multi-fast-field-reader rust:revert-1094-expose-min-max-on-multi-fast-field-reader rust:hotfix-1088 rust:owned-bytes-instead-of-rawdoc rust:rihardsk-date-ranges rust:issue/997 rust:0.13.3 rust:fail-fast-store-checkpoint rust:debugging-store rust:issue/reproduce-897 rust:issue/969-reproduce rust:troublescooter-static rust:intermediate-mergeable-troublescooter rust:barrotsteindev-filter-collector-tpredicatevalue rust:issue/938b rust:issue/938 rust:barrotsteindev-filter-collector rust:issue/922 rust:issue/896 rust:quickfix-explain rust:slog rust:limit-random-seeks rust:0.13.1 rust:issue/866b rust:bugfix-unittest-macos rust:githubactions rust:fieldnorm_readers rust:blockwand rust:nodeffeatfails rust:removedaliveiterator rust:robyoung-introduce-offset-to-top-docs rust:specialization-of-foreach-for-union-scorer rust:elshize-bmw rust:cutt rust:nrt rust:storewriter-out rust:segment_writer_in_ram2 rust:polling rust:audunhalland-query-boost rust:poll-meta rust:segment_writer_in_ram rust:bundle rust:petr-tik-n662_footer_is_incompatible_impl rust:race-condition-in-tests rust:hotfix-0.10.3-issue681 rust:remove-tru rust:kkoziara-pre-tokenized-text rust:issue/681 rust:issue/680 rust:refactoring/field rust:updating/uuid rust:fix-bench rust:hotfix-656 rust:readd-macos-ci rust:incremental-search rust:crate-division rust:streamer-with-state rust:criterion rust:python-binding-changes rust:bump-version rust:gh-pages rust:issue/554 rust:perf/experimental-union-for_each rust:issue/526b rust:softcommits rust:0.9.1-hotfix rust:revert-521-fix-non-english-stemmers rust:issue/500-hotfix rust:hotfix/0.8.2 rust:issue/483 rust:bug/merge-deted rust:issue/407 rust:issue/457 rust:segment_fruits rust:doc rust:dds/lenient rust:issue/webasm rust:issue/396 rust:broken_collector_with_result rust:dss/automaton-weight-suggestion rust:wasm rust:common-crawl rust:tantivy-imhotep
rust:0.25.0 rust:0.22.1 rust:0.24.2 rust:0.24.1 rust:tantivy-query-grammar-v0.24.0 rust:tantivy-columnar-v0.5.0 rust:tantivy-bitpacker-v0.8.0 rust:ownedbytes-v0.9.0 rust:tantivy-stacker-v0.5.0 rust:tantivy-sstable-v0.5.0 rust:tantivy-common-v0.9.0 rust:tantivy-tokenizer-api-v0.5.0 rust:0.24 rust:qw-airmail-20250224-hotfix-merge-panic rust:qw-airmail-20250219-hotfix-merge-panic rust:qw-airmail-202406 rust:qw-airmail-20240611 rust:qw-airmail-20240606 rust:quickwit-rev rust:0.22.0 rust:tantivy-stacker-v0.3.0 rust:ownedbytes-v0.7.0 rust:tantivy-tokenizer-api-v0.3.0 rust:tantivy-columnar-v0.3.0 rust:tantivy-common-v0.7.0 rust:tantivy-query-grammar-v0.22.0 rust:tantivy-sstable-v0.3.0 rust:tantivy-bitpacker-v0.6.0 rust:0.21.1 rust:0.21 rust:0.20.2 rust:0.20.1 rust:0.20 rust:quickwit-0.5-rev rust:0.19.2 rust:0.19.1 rust:0.19 rust:0.18.1 rust:0.18 rust:0.17 rust:0.16.1 rust:0.15.3 rust:0.15.2 rust:0.15.1 rust:0.15 rust:0.14 rust:0.13.3 rust:0.13.2 rust:0.13.1 rust:0.13 rust:0.12 rust:0.11.3 rust:0.11.1 rust:0.11 rust:0.10.3 rust:0.1 rust:0.10.2 rust:0.10.1 rust:0.10.0 rust:0.9.1 rust:0.9 rust:0.8.2 rust:0.8.2b rust:0.8.1 rust:0.8.0 rust:0.7.2 rust:0.7.1 rust:0.7.0 rust:0.6.1 rust:0.6.0 rust:0.5.2 rust:0.5.1 rust:0.5.0 rust:0.4.3 rust:0.4.2 rust:0.4.0 rust:0.3.1 rust:0.3.0 rust:0.2.0

437 Commits
0.2.0 ... 0.4.3

Author SHA1 Message Date
Paul Masurel
4f5ce12a77 NOBUG removed cpp from patterns 2018-01-05 12:09:42 +09:00
Paul Masurel
813efa4ab3 NOBUG coveralls 2018-01-05 11:03:27 +09:00
Paul Masurel
c3b6c1dc0b NOBUG coveralls 2018-01-05 00:31:57 +09:00
Paul Masurel
6f5e0ef6f4 NOBUG Simplify travis 2018-01-04 20:51:00 +09:00
Paul Masurel
7224f58895 Merge branch 'issue/218' 
Conflicts:
	src/directory/mmap_directory.rs
	src/lib.rs
 2018-01-04 18:47:10 +09:00
Paul Masurel
49519c3f61 added comments 2018-01-04 12:53:20 +09:00
Paul Masurel
cb11b92505 Added comments 2018-01-04 12:27:14 +09:00
Paul Masurel
7b2dcfbd91 Merge branch 'issue/227' 2018-01-04 12:12:00 +09:00
Paul Masurel
d2e30e6681 Merge branch 'master' of github.com:tantivy-search/tantivy 2018-01-04 12:09:44 +09:00
Paul Masurel
ef109927b3 rustfmt 2018-01-04 12:08:34 +09:00
Paul Masurel
44e5c4dfd3 Added alphanum only token filter 2017-12-31 13:43:10 +09:00
Paul Masurel
6f223253ea Made load_metas public 2017-12-31 08:57:19 +09:00
Paul Masurel
f7b0392bd5 issue/230 Add an optional commit message. (#231) 
Closes #230
 2017-12-27 12:27:02 +09:00
Paul Masurel
442bc9a1b8 Fixes the computation of the memory size of a hashtable with a key of n bits. (#229) 
Closes #228
 2017-12-25 13:04:10 +09:00
Paul Masurel
db7d784573 Issue 227 Faster merge when there are no deletes 2017-12-21 22:04:05 +09:00
Paul Masurel
79132e803a NOBUG Switched to 64 bits addr 2017-12-21 11:06:46 +09:00
Paul Masurel
9e132b7dde NOBUG QueryParser does not need to be mut. Code cleanup 2017-12-16 15:43:35 +09:00
Paul Masurel
1e55189db1 NOBUG rustfmt 2017-12-14 19:30:31 +09:00
Paul Masurel
8b1b389a76 NOBUG Clippy 2017-12-14 19:25:12 +09:00
Paul Masurel
46f3ec87a5 Removed packed memory layout. 2017-12-14 18:37:04 +09:00
Paul Masurel
f24e5f405e NOBUG intellij misc lint 2017-12-14 18:23:35 +09:00
Paul Masurel
2589be3984 BUGFIX Serialization of schema got broken after serde's update 2017-12-14 17:37:20 +09:00
Paul Masurel
a02a9294e4 removed doc in travis 2017-11-27 13:53:58 +09:00
Paul Masurel
8023445b63 docs 2017-11-26 11:52:03 +09:00
Paul Masurel
05ce093f97 doc 2017-11-26 11:43:11 +09:00
Paul Masurel
6937e23a56 fixing doctest 2017-11-26 11:06:34 +09:00
Paul Masurel
974c321153 cargo fmt 2017-11-26 11:02:02 +09:00
Paul Masurel
f30ec9b36b Merge branch 'master' of github.com:tantivy-search/tantivy 
Conflicts:
	src/analyzer/mod.rs
	src/schema/index_record_option.rs
	src/tokenizer/lower_caser.rs
	src/tokenizer/tokenizer.rs
 2017-11-26 10:54:05 +09:00
Paul Masurel
acd7c1ea2d Added comments 2017-11-26 10:44:49 +09:00
Paul Masurel
aaeeda2bc5 Editing rustdoc 2017-11-25 13:23:32 +09:00
Paul Masurel
ac4d433fad Renamed analyzer to tokenizer 2017-11-24 16:50:32 +09:00
Paul Masurel
a298c084e6 Analyzer's Analyzer::token_stream does not need to me &mut self 2017-11-22 20:37:34 +09:00
Paul Masurel
185a72b341 Closes #224. Fixes documentation about STORED in the example. (#225) 2017-11-16 08:22:54 +09:00
Paul Masurel
bb41ae76f9 Closes #224. Fixes documentation about STORED in the example. 2017-11-16 08:16:17 +09:00
Paul Masurel
74d32e522a Stopped using mmap in tantivy. Caching MmapReadOnly. 
Closes #218
 2017-10-08 17:07:19 +09:00
Jain Jacob
927dd1ee6f Updates crate gcc to cc v1 (#217) 
* Bump cc to v1

* Changes gcc::Config to cc::Build. Resolves #216
 2017-10-06 16:18:44 +09:00
Paul Masurel
2c9302290f #191 Analyzer 2017-09-20 22:56:55 +09:00
Paul Masurel
426cc436da Test passing 2017-09-10 17:48:41 +09:00
Paul Masurel
68d42c9cf2 Added raw tokenizer, using the right analyzer in query parser. 2017-09-10 16:58:50 +09:00
Paul Masurel
ca49d6130f Test not passing 2017-09-09 17:32:47 +09:00
Paul Masurel
3588ca0561 Integrated with the merge branch 2017-09-09 15:27:19 +09:00
Paul Masurel
7c6cdcd876 Merge branch 'master' of github.com:tantivy-search/tantivy 2017-09-02 16:03:06 +09:00
Paul Masurel
71366b9a56 issue/197 Remove logic that prevents leak from crossbeam MsQueue. (#212) 
Closes #197
 2017-09-02 15:55:23 +09:00
Paul Masurel
a3247ebcfb issue/197 Remove logic that prevents leak from csossbeam MsQueue. 2017-09-02 15:53:07 +09:00
Paul Masurel
3ec13a8719 Readded fix for non-simd 2017-08-28 23:18:56 +09:00
Paul Masurel
f8593c76d5 Merge branch 'imhotep-new-codec' 
Conflicts:
	src/common/bitpacker.rs
	src/compression/pack/compression_pack_nosimd.rs
	src/indexer/log_merge_policy.rs
 2017-08-28 19:30:01 +09:00
Paul Masurel
f8710bd4b0 Format 2017-08-28 18:22:41 +09:00
Paul Masurel
8d05b8f7b2 Added comments. Renamed field reader 2017-08-28 17:00:12 +09:00
Paul Masurel
fc25516b7a Added unit test. 2017-08-28 11:15:37 +09:00
Paul Masurel
5b1e71947f Stream working, all test passing 2017-08-27 20:20:38 +09:00
Paul Masurel
69351fb4a5 Toward a new codec 2017-08-27 18:44:37 +09:00
Paul Masurel
3d0082d020 Delta encoded. Range and get are broken 2017-08-26 19:59:51 +09:00
Paul Masurel
8e450c770a Better error handling. Some doc. 2017-08-26 18:40:30 +09:00
Paul Masurel
a757902aed Merge branch 'feature/streamdict-simd' into imhotep 2017-08-22 18:58:57 +09:00
Paul Masurel
b3a8074826 removed println 2017-08-22 18:58:17 +09:00
Paul Masurel
4289625348 Merged with the new codec branch 2017-08-22 18:26:09 +09:00
Paul Masurel
850f10c1fe Exposing Field 2017-08-22 18:21:35 +09:00
raphael claude
d7f9bfdfc5 fix segments sorting in log_merge_policy (#211) 
bug: segments were sorted on their indices (first field in the tuples)
fix: sort on the segments size
 2017-08-20 08:59:54 +09:00
Paul Masurel
d0d5db4515 Streamdict using SIMD instruction. 2017-08-19 12:03:04 +09:00
Paul Masurel
303fc7e820 Better unit test for termdict. Checking the TermInfo 2017-08-17 12:08:39 +09:00
Paul Masurel
744edb2c5c NOBUG Avoid serializing position offset when useless. Test passing 2017-08-16 14:06:00 +09:00
Paul Masurel
2d70efb7b0 Removed trait boundary on termdict 2017-08-15 14:43:05 +09:00
Paul Masurel
eb5b2ffdcc Cleanups 2017-08-15 13:57:22 +09:00
Paul Masurel
38513014d5 Reenable unit test. 
Consuming CompositeWrite on Close.
 2017-08-14 23:35:09 +09:00
Paul Masurel
9cb7a0f6e6 Unit tests passing 2017-08-13 19:38:25 +09:00
Paul Masurel
8d466b8a76 half way through removing FastFieldsReader 2017-08-13 18:39:45 +09:00
Paul Masurel
413d0e1719 NOBUG test passing 2017-08-13 17:57:11 +09:00
Paul Masurel
0eb3c872fd Using composite file for all of the inverted index component 2017-08-12 19:34:23 +09:00
Paul Masurel
f9203228be Using composite file in fast field. 2017-08-12 18:45:59 +09:00
Paul Masurel
8f377b92d0 introducing a field serializer 2017-08-11 18:11:32 +09:00
Paul Masurel
1e89f86267 blop 2017-08-08 13:55:09 +09:00
Paul Masurel
d1f61a50c1 issue/207 Lazily decompressing positions. 2017-08-06 20:29:21 +09:00
Dru Sellers
2bb85ed575 Minor Doc Changes (#206) 
* Various small documentation tweaks

* walking through the docs

* Update lib.rs

* Update lib.rs

* Update mod.rs
 2017-08-06 09:22:03 +09:00
Paul Masurel
236fa74767 Positions almost working. 2017-08-05 23:17:35 +09:00
Paul Masurel
63b35dd87b removing freq handler. 2017-08-05 18:09:19 +09:00
Paul Masurel
efb910f4e8 Added CompressedIntStream 2017-08-05 16:44:01 +09:00
Paul Masurel
aff7e64d4e test 2017-08-04 22:07:14 +09:00
Paul Masurel
92a3f3981f issue/204 trying to fix nosimd branch. test not passing 2017-08-04 21:19:18 +09:00
king6cong
447a9361d8 Remove submodule information in README as subtree is now used 2017-08-03 13:52:16 +09:00
Paul Masurel
5f59139484 NOBUG simplified code. 2017-08-02 20:49:47 +09:00
Paul Masurel
27c373d26d NOBUG Updated changelog and bumped version 2017-07-24 18:52:45 +09:00
Paul Masurel
80ae136646 issue/198 Getting living_file after getting the list of managed files. 2017-07-24 18:46:41 +09:00
Paul Masurel
52b1398702 NOBUG version 0.4.0 -> 0.4.1 2017-07-19 19:07:54 +09:00
Paul Masurel
7b9cd09a6e Closes #199. Unindexed fields are indexed as untokenized 2017-07-19 18:41:22 +09:00
Paul Masurel
4c423ad2ca Merge branch 'master' of github.com:tantivy-search/tantivy 2017-07-19 17:01:32 +09:00
Paul Masurel
9f542d5252 NOBUG Fix spelling of "encountered". (as reported by @dazzag24) 2017-07-19 16:59:50 +09:00
Paul Masurel
77d8e81ae4 issue/17 Slightly more explicit error message 2017-07-19 11:08:42 +09:00
Paul Masurel
76e07b9705 NOBUG Small fixes. 2017-07-14 18:09:54 +09:00
Paul Masurel
ea4e9fdaf1 NOBUG updated README 2017-07-14 14:09:13 +09:00
Paul Masurel
e418bee693 NOBUG Garbage collection after end merge. 2017-07-14 12:09:47 +09:00
Paul Masurel
af4f1a86bc Merge remote-tracking branch 'origin/exp/hash_intable' 2017-07-13 20:50:54 +09:00
Paul Masurel
753b639454 NOBUG splitting the per-thread memory between the table and the heap 2017-07-13 17:11:39 +09:00
Paul Masurel
5907a47547 NOBUG Added whitespaces. 2017-07-13 15:14:12 +09:00
Paul Masurel
586a6e62a2 NOBUG Added Changelog for 4.0 2017-07-13 15:06:09 +09:00
Paul Masurel
fdae0eff5a NOBUG Remove range step_by 2017-07-13 14:05:33 +09:00
Paul Masurel
6eea407f20 Removing usage of step_by 2017-06-23 17:46:39 +09:00
Paul Masurel
1ba51d4dc4 NOBUG removed using range.step_by 2017-06-22 22:10:53 +09:00
Paul Masurel
6e742d5145 NOBUG removing batch add docs 2017-06-22 11:35:22 +09:00
Paul Masurel
1843259e91 NOBUG Simplified addr definitions 2017-06-22 11:27:32 +09:00
Paul Masurel
4ebacb7297 BytesRef is now wrapping an addr 2017-06-21 22:32:05 +09:00
Paul Masurel
fb75e60c6e issue/136 Added hashmaps. 2017-06-21 15:47:55 +09:00
Paul Masurel
04b15c6c11 Merge branch 'master' into exp/hash_intable 
Conflicts:
	src/datastruct/stacker/hashmap.rs
 2017-06-21 11:40:49 +09:00
Paul Masurel
b05b5f5487 issue/191 Added an analyzer manager. 2017-06-20 10:02:26 +09:00
Paul Masurel
4fe96483bc fill_buffer 2017-06-14 23:32:58 +09:00
Paul Masurel
09e27740e2 Added fill_buffer in DocSet 2017-06-14 18:28:30 +09:00
Paul Masurel
e51feea574 Removed cargo fmt from travis. 2017-06-14 13:45:11 +09:00
Paul Masurel
93e7f28cc0 Added unit test 2017-06-14 10:46:06 +09:00
Paul Masurel
8875b9794a Added API to get range from fastfield 2017-06-13 23:16:50 +09:00
Paul Masurel
f26874557e Remove the concept of pipeline. Made a BoableAnalyzer 2017-06-10 20:06:00 +09:00
Paul Masurel
a7d10b65ae Added support for Japanese. 2017-06-09 22:25:03 +09:00
Paul Masurel
e120e3b7aa issue/191 Added proper analyzer 2017-06-07 23:21:36 +09:00
Paul Masurel
90fcfb3f43 issue/188 Using murmurhash 2017-06-07 09:30:34 +09:00
Paul Masurel
e547e8abad Closes #184 
Resizing the `Vec` was a bad idea, as for some stacker operation,
we may have a living reference to an object in the current heap.
 2017-06-06 23:16:28 +09:00
Paul Masurel
5aa4565424 Tiny cleaning 2017-06-05 23:40:08 +09:00
Paul Masurel
3637620187 Merge branch 'master' of github.com:tantivy-search/tantivy 2017-06-02 21:03:37 +09:00
Laurentiu Nicola
a94679d74d Use four terms in the intersection bench 2017-05-31 08:31:33 +09:00
Laurentiu Nicola
a35a8638cc Comment nit 2017-05-31 08:31:33 +09:00
Paul Masurel
97a051996f issue 171. Hopefully bugfix? 2017-05-31 08:31:33 +09:00
Laurentiu Nicola
69525cb3c7 Add extra intersection test 2017-05-31 08:31:33 +09:00
Laurentiu Nicola
63867a7150 Fix document generation for posting benchmarks 2017-05-31 08:31:33 +09:00
Paul Masurel
19c073385a Better intersection and added size_hint 2017-05-31 08:31:33 +09:00
Paul Masurel
0521844e56 Format, small changes in VInt 2017-05-31 08:31:20 +09:00
Paul Masurel
8d4778f94d issue/181 BinarySerializable does not return the len + Generics over Read+Write 2017-05-31 08:31:20 +09:00
Paul Masurel
1d5464351d generic read 2017-05-31 08:31:20 +09:00
Paul Masurel
522ebdc674 made ResultExt public 2017-05-31 08:31:20 +09:00
Paul Masurel
4a805733db another hash 2017-05-30 15:36:48 +09:00
Paul Masurel
568d149db8 Merge branch 'master' into exp/hash_intable 2017-05-30 08:27:33 +09:00
Paul Masurel
4cfc9806c0 made ResultExt public 2017-05-30 08:22:17 +09:00
Paul Masurel
37042e3ccb Send and Sync impl now useless 2017-05-29 18:53:49 +09:00
Paul Masurel
b316cd337a Optimization in bitpacker 2017-05-29 18:53:49 +09:00
Paul Masurel
c04991e5ad Removed pointer in fastfield 2017-05-29 18:53:49 +09:00
Paul Masurel
c59b712eeb Added hash info in the table 2017-05-29 18:47:20 +09:00
Ashley Mannix
da61baed3b run fmt 2017-05-29 18:29:39 +09:00
Ashley Mannix
b6140d2962 drop some patch bounds 2017-05-29 18:29:39 +09:00
Ashley Mannix
6a9a71bb1b re-export ErrorKind 2017-05-29 18:29:39 +09:00
Ashley Mannix
e8fc4c77e2 fix delete error msg 2017-05-29 18:29:39 +09:00
Ashley Mannix
80837601ea remove error::* imports 2017-05-29 18:29:39 +09:00
Ashley Mannix
2b2703cf51 run cargo fmt 2017-05-29 18:29:39 +09:00
Ashley Mannix
d79018a7f8 fix build warnings 2017-05-29 18:29:39 +09:00
Ashley Mannix
d8a7c428f7 impl std error for directory errors 2017-05-29 18:29:39 +09:00
Ashley Mannix
45595234cc fix error match 2017-05-29 18:29:39 +09:00
Ashley Mannix
1bcebdd29e initial error-chain 2017-05-29 18:29:39 +09:00
Paul Masurel
ed0333a404 Optimized streamer 2017-05-28 19:58:28 +09:00
Paul Masurel
ac0b1a21eb Term as a wrapper 
Small changes

Plastic
 2017-05-25 23:49:54 +09:00
Paul Masurel
6bbc789d84 Fmt fix 2017-05-25 23:49:54 +09:00
Paul Masurel
87152daef3 issue/174 Added doc, and made field private 2017-05-25 23:49:54 +09:00
Paul Masurel
e0fce4782a Added documentation 2017-05-25 23:49:54 +09:00
Paul Masurel
a633c2a49a Avoid exposing common. Exposes u64 to i64 conversion instead. 2017-05-25 23:49:54 +09:00
Paul Masurel
51623d593e Avoid exposign schema from segment_reader 2017-05-25 23:49:54 +09:00
Paul Masurel
29bf740ddf Exposing the remaining API 2017-05-25 23:49:54 +09:00
Paul Masurel
511bd25a31 trailing whitespace 2017-05-25 18:17:37 +09:00
Paul Masurel
66e14ac1b1 clippy 2017-05-25 18:17:37 +09:00
Paul Masurel
09e94072ba Cargo fmt 2017-05-25 18:17:37 +09:00
Paul Masurel
6c68136d31 Reorganized code 2017-05-25 18:17:37 +09:00
Paul Masurel
aaf1b2c6b6 Reorganized code and added documentation. 2017-05-25 18:17:37 +09:00
Paul Masurel
8a6af2aefa Added unit test and bugfix 2017-05-25 18:17:37 +09:00
Paul Masurel
7a6e62976b Added stream dictionary code, merge unit test 2017-05-25 18:17:37 +09:00
Paul Masurel
2712930bd6 Added the feature 2017-05-25 18:17:37 +09:00
Paul Masurel
cb05f8c098 Prevent execution of the code in the macro doc 2017-05-22 10:55:45 +09:00
Paul Masurel
c0c9d04ca9 Added extra doc 2017-05-22 10:55:45 +09:00
Paul Masurel
7ea5e740e0 Using the $crate thing to make the macro usable in and outside tantivy 2017-05-22 10:55:45 +09:00
Paul Masurel
2afa6c372a issue/168 Make doc! macro usable outside tantivy 2017-05-22 10:55:45 +09:00
Paul Masurel
c7db8866b5 Merge branch 'facets' 2017-05-21 22:57:01 +09:00
Paul Masurel
02d992324a simplified facets. 2017-05-21 22:56:43 +09:00
Paul Masurel
4ab511ffc6 Merging 2017-05-21 22:15:02 +09:00
Paul Masurel
f318172ea4 Merge branch 'issue/162' 2017-05-21 20:04:03 +09:00
Paul Masurel
581449a824 issue/162 Docs and unit tests 2017-05-21 18:58:04 +09:00
Maciej Dziardziel
272589a381 faceting for fast numerical fields 2017-05-21 12:04:29 +03:00
Laurentiu Nicola
73d54c6379 Inline block_len 2017-05-21 10:44:49 +03:00
Paul Masurel
3e4606de5d Simplifying, and reordering the members 2017-05-21 16:31:52 +09:00
Laurentiu Nicola
020779f61b Make things faster 2017-05-20 20:56:37 +03:00
Laurentiu Nicola
835936585f Don't search whole blocks, but only the remaining part 2017-05-20 18:45:41 +03:00
Paul Masurel
bdd05e97d1 Added bench for segment postings 2017-05-20 23:38:53 +09:00
Paul Masurel
2be5f08cd6 issue/162 Added block iteration API 2017-05-20 11:46:40 +09:00
Paul Masurel
3f49d65a87 issue/162 Create block postings 2017-05-20 00:46:23 +09:00
Paul Masurel
f9baf4bcc8 Merge branch 'issue/155' 
Conflicts:
	src/indexer/merger.rs
	src/indexer/segment_writer.rs
 2017-05-19 20:14:36 +09:00
Paul Masurel
7ee93fbed5 Cleaning 2017-05-19 20:08:04 +09:00
Paul Masurel
57a5547ae8 Comments and cleaning up API 2017-05-19 11:20:27 +09:00
Paul Masurel
c57ab6a335 Renamed fstmap to termdict 2017-05-19 09:26:18 +09:00
Paul Masurel
02bfa9be52 Moving to termdict 2017-05-19 08:43:52 +09:00
Paul Masurel
b3f62b8acc Better API 2017-05-18 23:35:39 +09:00
Paul Masurel
2a08c247af Clippy 2017-05-18 23:20:41 +09:00
Paul Masurel
d2926b6ee0 Format 2017-05-18 23:09:20 +09:00
Paul Masurel
0272167c2e Code cleaning 2017-05-18 23:06:02 +09:00
Laurentiu Nicola
a9cf0bde16 Format code 2017-05-18 22:07:49 +09:00
Laurentiu Nicola
5a457df45d VInt encode values in IntFastFieldWriter 
Closes #131
 2017-05-18 22:07:49 +09:00
Paul Masurel
ca76fd5ba0 Uncommenting unit test 2017-05-18 20:41:56 +09:00
Paul Masurel
e79a316e41 Issue 155 - Trying to avoid term lookup when merging terms 
+ Adds a proper Streamer interface
 2017-05-18 20:12:00 +09:00
Paul Masurel
733f54d80e Making clippy happy. 2017-05-17 19:07:39 +09:00
Paul Masurel
7b2b181652 Merge branch 'master' into issue/136 
Conflicts:
	src/datastruct/stacker/hashmap.rs
	src/datastruct/stacker/heap.rs
	src/datastruct/stacker/mod.rs
	src/indexer/index_writer.rs
	src/indexer/merger.rs
	src/indexer/segment_updater.rs
	src/indexer/segment_writer.rs
	src/postings/postings_writer.rs
	src/postings/recorder.rs
	src/schema/term.rs
 2017-05-17 18:40:09 +09:00
Laurentiu Nicola
b3f39f2343 Remove unneeded suppressions, make clippy lints explicit 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
a13122d392 use explicit drop instead of suppression 2017-05-17 15:50:07 +09:00
Paul Masurel
113917c521 Making clippy happy. 
+ Simplifying bitpacking by adding a 7 byte padding.
+ Bugfix in a unit test.
 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
1352b95b07 clippy: fix never_loop warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
c0538dbe9a clippy: fix mut_from_ref warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
0d5ea98132 clippy: fix inline_always warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
0404df3fd5 Fix typo in docstring 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
a67caee141 clippy: fix len_zero warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
f5fb29422a clippy: fix while_let_loop warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
4e48bbf0ea clippy: fix needless_lifetimes warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
6fea510869 clippy: fix redundant_closure warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
39958ec476 clippy: fix single_match warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
36f51e289e clippy: fix match_same_arms warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
5c83153035 clippy: fix or_fun_call warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
8e407bb314 clippy: fix needless_borrow warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
103ba6ba35 clippy: fix match_ref_pats warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
3965b26cd2 clippy: fix useless_let_if_seq warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
1cd0b378fb clippy: fix map_clone warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
92f383fa51 clippy: fix let_unit_value warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
6ae34d2a77 clippy: fix toplevel_ref_arg warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
1af1f7e0d1 clippy: fix if_let_redundant_pattern_matching warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
feec2e2620 clippy: fix needless_bool warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
3e2ad7542d clippy: fix needless_return warnings 2017-05-17 15:50:07 +09:00
Laurentiu Nicola
ac02c76b1e clippy: fix doc_markdown warnings 2017-05-17 15:50:07 +09:00
Paul Masurel
e5c7c0b8b9 Update CHANGELOG.md 2017-05-16 21:13:33 +09:00
Laurentiu Nicola
49dbe4722f Add a test for SegmentPostings::skip_len 2017-05-16 21:12:43 +09:00
Laurentiu Nicola
f64ff77424 Use an exponential search 2017-05-16 21:12:43 +09:00
Laurentiu Nicola
2bf93e9e51 Avoid rebuilding simdcomp when running tests 2017-05-16 08:37:43 +09:00
Laurentiu Nicola
3dde748b25 Make rustfmt happy 2017-05-16 00:49:05 +03:00
Laurentiu Nicola
1dabe26395 Add comment about block_len 2017-05-15 21:26:28 +03:00
Laurentiu Nicola
5590537739 Disable early exit 2017-05-15 21:18:06 +03:00
Laurentiu Nicola
ccf0f9cb2f Merge branch 'master' of github.com:tantivy-search/tantivy into issue/130 2017-05-15 18:54:16 +03:00
Laurentiu Nicola
e21913ecdc Use binary search for SegmentPostings::skip_next 2017-05-15 18:33:43 +03:00
Laurentiu Nicola
2cc826adc7 Add a bench for SegmentPostings::SkipNext 2017-05-15 18:33:43 +03:00
Laurentiu Nicola
4d90d8fc1d Move the random sampling helpers to the tests module 2017-05-15 18:33:43 +03:00
Paul Masurel
0606a8ae73 Bugfix in travis yml 2017-05-16 00:22:11 +09:00
Paul Masurel
03564214e7 Added check for rustfmt in travis 2017-05-15 22:46:43 +09:00
Paul Masurel
4c8f9742f8 format 2017-05-15 22:30:18 +09:00
Paul Masurel
a23b7a1815 Test the size of complete 0..128 block 2017-05-15 19:09:52 +09:00
Paul Masurel
6f89a86b14 Added simple search in travis CI 2017-05-15 12:10:23 +09:00
Laurentiu Nicola
b2beac1203 Check the result of wait_merging_threads 2017-05-15 08:00:25 +09:00
Paul Masurel
8cd5a2d81d Fixed logging deleted files twice 2017-05-15 00:25:49 +09:00
Paul Masurel
b26c22ada0 Merge branch 'issue/148' 2017-05-15 00:02:51 +09:00
Laurentiu Nicola
8a35259300 Avoid clone() call 2017-05-14 23:28:17 +09:00
Paul Masurel
db56167a5d Display backtrace 2017-05-14 23:28:17 +09:00
Paul Masurel
ab66ffed4e Closes #147 2017-05-14 23:28:17 +09:00
Laurentiu Nicola
e04f2f0b08 issue/148 Wait for the index writer threads to shut down in simple_search 2017-05-14 16:35:24 +03:00
Paul Masurel
7a5df33c85 issue/148 Wrapping MsQueue to drop all of its concent on Drop 2017-05-14 16:25:33 +03:00
Laurentiu Nicola
ee0873dd07 Avoid clone() call 2017-05-13 16:11:58 +03:00
Paul Masurel
695c8828b8 Display backtrace 2017-05-13 18:51:38 +09:00
Paul Masurel
4ff7dc7a4f Closes #147 2017-05-13 18:46:50 +09:00
Paul Masurel
69832bfd03 NOBUG Disabling running examples in CI as it is not working. 2017-05-12 14:35:50 +09:00
Paul Masurel
ecbdd70c37 Removed the clunky linked list logic of the heap. 2017-05-12 14:01:52 +09:00
Paul Masurel
fb1b2be782 issue/136 Fix following CR 2017-05-12 13:51:09 +09:00
Paul Masurel
9cd7458978 NOBUG Hiding methods making it possible to build a incorrect Term. 2017-05-11 21:12:59 +09:00
Paul Masurel
4c4c28e2c4 Fix broke compile 2017-05-11 20:57:32 +09:00
Paul Masurel
9f9e588905 Merge branch 'master' into issue/136 
Conflicts:
	src/postings/postings_writer.rs
 2017-05-11 20:50:24 +09:00
Paul Masurel
6fd17e0ead Code cleaning 2017-05-11 20:47:30 +09:00
Paul Masurel
65dc5b0d83 Closes #145 2017-05-11 19:48:06 +09:00
Paul Masurel
15d15c01f8 Runing examples in CI 
Closes #143
 2017-05-11 19:43:36 +09:00
Paul Masurel
106832a66a Make Term::with_capacity crate-public 2017-05-11 19:37:15 +09:00
Paul Masurel
477b9136b9 FIXED inconsistent Term's field serialization. 
Also.

Cleaned up the code to make sure that the logic
is only in one place.
Removed allocate_vec

Closes #141
Closes #139
Closes #142
Closes #138
 2017-05-11 19:37:15 +09:00
Paul Masurel
7852d097b8 CHANGELOG 0.3.1 did not included the fix of the Field(u32) 2017-05-11 09:48:37 +09:00
Ashley Mannix
0bd56241bb pretty print meta.json 2017-05-10 20:13:53 +09:00
Paul Masurel
54ab897755 Added comment 2017-05-10 19:30:24 +09:00
Paul Masurel
1369d2d144 Quadratic probing. 2017-05-10 10:38:47 +09:00
Paul Masurel
d3f829dc8a Bugfix 2017-05-10 00:29:37 +09:00
Paul Masurel
e82ccf9627 Merge branch 'master' into issue/indexing-refactoring 2017-05-09 16:43:33 +09:00
Paul Masurel
d3d29f7f54 NOBUG Updated CHANGELOG with the serde change for 0.4.0 2017-05-09 16:42:25 +09:00
Paul Masurel
3566717979 Merge pull request #134 from tantivy-search/chore/serde-rebase 
Replace rustc_serialize with serde (updated)
 2017-05-09 16:38:42 +09:00
Paul Masurel
90bc3e3773 Added limitation on term dictionary saturation 2017-05-09 14:10:33 +09:00
Paul Masurel
ffb62b6835 working 2017-05-09 10:17:05 +09:00
Ashley Mannix
4f9ce91d6a update underflow test 2017-05-08 14:40:58 +10:00
Laurentiu Nicola
3c3a2fbfe8 Remove old serialization code 2017-05-08 07:36:15 +03:00
Laurentiu Nicola
0508571d1a Use the proper error type on u64 overflow 2017-05-08 07:35:33 +03:00
Laurentiu Nicola
7b733dd34f Fix i64 overflow check and merge NotJSON with NotJSONObject 2017-05-08 07:09:54 +03:00
Ashley Mannix
2c798e3147 Replace rustc_serialize with serde 2017-05-07 20:21:22 +03:00
Paul Masurel
2c13f210bc Bugfix on merging i64 fast fields 2017-05-07 15:57:29 +09:00
Paul Masurel
0dad02791c issues/65 Added comments 
Closes #65
Closes #132
 2017-05-06 23:09:45 +09:00
Paul Masurel
2947364ae1 issues/65 Phrase query for untokenized fields are not tokenized. 2017-05-06 22:14:26 +09:00
Paul Masurel
05111599b3 Removed several TODOs 2017-05-05 16:08:09 +08:00
Paul Masurel
83263eabbb issues/65 Updated changelog added some doc. 2017-05-04 17:13:14 +08:00
Paul Masurel
5cb5c9a8f2 issues/65 Added i64 fast fields 2017-05-04 16:46:14 +08:00
Paul Masurel
9ab92b7739 i64 fast field working 2017-05-04 16:46:14 +08:00
Paul Masurel
962bddfbbf Merge with panicks. 2017-05-04 16:46:14 +08:00
Paul Masurel
26cfe2909f FastField with different types 2017-05-04 16:46:13 +08:00
Paul Masurel
afdfb1a69b Compiling... fastfield not implemented yet 2017-05-04 16:46:13 +08:00
Paul Masurel
b26ad1d57a Added int options 2017-05-04 16:46:13 +08:00
Paul Masurel
1dbd54edbb Renamed u64options 2017-05-04 16:46:13 +08:00
Paul Masurel
deb04eb090 issue/65 Switching to u64. 2017-05-04 16:46:13 +08:00
Paul Masurel
bed34bf502 Merge branch 'issues/122' 2017-04-23 16:14:40 +08:00
Paul Masurel
80f1e26c3b Tantivy 0.3.1 2017-04-23 15:52:07 +08:00
Paul Masurel
3e68b61d8f issue/122 Adds a garbage collect method 2017-04-23 15:51:06 +08:00
Paul Masurel
95bfb71901 NOBUG Remove 256 num fields limit 2017-04-19 22:37:34 +09:00
Paul Masurel
74e10843a7 issue/120 Disabled SIMD vbyte compression for msvc 2017-04-17 22:36:32 +09:00
Paul Masurel
1b922e6d23 issue 120. Using streamvbyte codec for the vbyte part of the encoding 2017-04-16 18:49:53 +09:00
Paul Masurel
a7c6c31538 Merge commit '9d071c8d4610aa61f4b1f7dd489210415a05cfc0' as 'cpp/streamvbyte' 2017-04-16 15:22:43 +09:00
Paul Masurel
9d071c8d46 Squashed 'cpp/streamvbyte/' content from commit f38aa6b 
git-subtree-dir: cpp/streamvbyte
git-subtree-split: f38aa6b6ec4c5cee9d72c94ef305e6a79a108252
 2017-04-16 15:22:43 +09:00
Paul Masurel
04074f7bcb Merge pull request #119 from tantivy-search/issue/118 
Using u32 for field ids
 2017-04-15 13:11:22 +09:00
Paul Masurel
8a28d1643d Using u32 for field ids 2017-04-15 13:04:33 +09:00
Paul Masurel
44c684af5c NOBUG Fixes winapi version 2017-04-08 19:01:31 +09:00
Paul Masurel
60279a03b6 RELEASE Tantivy 0.3. See Changelog 2017-04-08 18:53:40 +09:00
Paul Masurel
dc43135fe0 NOBUG Remove .info 2017-04-08 18:49:37 +09:00
Paul Masurel
ce022e5f06 issue/54 Clone segment reader rather than reload. 
Closes #54.
 2017-04-08 17:52:33 +09:00
Paul Masurel
0be977d9eb Merge pull request #114 from tantivy-search/issue/96 
Closes Issue/96
 2017-04-08 17:49:48 +09:00
Paul Masurel
a4ba20eea3 issue/96 code clean up, adding comments.wq 2017-04-08 17:30:25 +09:00
Paul Masurel
4bef6c99ee issue/96 Cleaning up some lock management 2017-04-05 10:12:39 +09:00
Paul Masurel
a84871468b issue/96 Rename FileError -> OpenReadError 2017-04-05 10:01:49 +09:00
Paul Masurel
e0a39fb273 issue/96 Added unit test, documentation and various tiny improvements. 2017-04-04 22:43:35 +09:00
Paul Masurel
35203378ef Considering merge options after calling end_merge 2017-04-03 17:26:21 +09:00
Paul Masurel
b5bf9bb13c issue/96 Looping over wait_merging_thread. 2017-04-03 08:39:18 +09:00
Paul Masurel
ea3349644c issue/96 Fixed unit test condition to something reasonable 2017-04-02 21:58:38 +09:00
Paul Masurel
d4f2e475ff issue/96 removed faulty assert 2017-04-02 19:21:20 +09:00
Paul Masurel
17631ed866 issue/96 Added functionality to protect files from deletion 
Hopefully fixed the race condition happening when merging files.
 2017-04-02 18:48:20 +09:00
Paul Masurel
9eb2d3e8c5 issue/96 avoid removing the bitset from segment_entry. 2017-04-02 16:26:28 +09:00
Paul Masurel
afd08a7bbc issue/96 Changed datastruct for the delete queue. 2017-04-01 21:01:10 +09:00
Paul Masurel
4fc7bc5f09 Added helper to create Vec with a given sizewq 2017-03-31 18:54:23 +09:00
Paul Masurel
602b9d235f Merge pull request #113 from kaedroho/patch-1 
Mark "cpp" folder as linguist-vendored in .gitattributes
 2017-03-31 09:05:57 +09:00
Karl Hobley
b22c6b86c7 Mark "cpp" folder as linguist-vendored in .gitattributes 
This repo is currently being detected as a C project because of some vendored libraries in the "cpp" folder.

According to https://github.com/github/linguist#using-gitattributes you can use ``.gitattributes`` tell GitHub to not count this folder when detecting the language.
 2017-03-30 13:43:03 +01:00
Paul Masurel
f0dc0de4b7 Added helper to create Vec with a given size 2017-03-29 11:26:24 +09:00
Paul Masurel
456dd3a60d issue/96 merge 2017-03-28 16:49:48 +09:00
Paul Masurel
d768a10077 master merged in feature branch 2017-03-27 09:27:23 +09:00
Paul Masurel
ddb2b8d807 test passing. 
SegmentWriter create SegmentEntry which contain a delete_bitset
 2017-03-26 18:32:53 +09:00
Paul Masurel
45806951b1 added quotation mark 2017-03-25 22:48:07 +09:00
Paul Masurel
84a060552d issue/109 trying to get proper logging in appveyor 2017-03-25 22:34:40 +09:00
Paul Masurel
68a956c6e7 issue/109 Showing debug! if test fails 2017-03-25 21:54:17 +09:00
Paul Masurel
f50f557cfc issue/109 Remove futures from most of segment_updater API. 2017-03-25 19:36:03 +09:00
Paul Masurel
daa19b770a (hopefully) bugfix race condition on wait merging threadwq. 2017-03-24 18:20:58 +09:00
Paul Masurel
e75402be80 Merge pull request #108 from KodrAus/ci/appveyor 
Add appveyor config
 2017-03-24 15:49:50 +09:00
Ashley Mannix
51cab39186 drop to vs2015 image 2017-03-24 16:37:30 +10:00
Ashley Mannix
c8e12b6847 try set mingw path 2017-03-24 16:22:32 +10:00
Ashley Mannix
b44a9cb89d add appveyor config 2017-03-24 16:11:51 +10:00
Paul Masurel
e650fab927 Merge pull request #106 from tantivy-search/wip/delay-test-deletes 
Fix delete tests on Windows
 2017-03-22 09:26:36 +09:00
Paul Masurel
b12a97abe4 Add unit test for when deleting fails 
Test that when delete fails, we still keep
the file as managed.

Remove the error log for windows, as failing
to delete is expected.
 2017-03-22 08:57:09 +09:00
Laurentiu Nicola
2b5a4bbde2 Don't delete twice on not(windows) 2017-03-21 07:48:58 +02:00
Laurentiu Nicola
2d169c4454 Delay deleting the files in the test suite to make it work on Windows 2017-03-21 07:37:28 +02:00
Paul Masurel
66d6e4e246 Merge pull request #103 from tantivy-search/lnicola-fix-sync-directory 
Make directory syncing work on Windows (resubmit)
 2017-03-21 10:55:03 +09:00
Paul Masurel
a061ba091d Merge pull request #105 from tantivy-search/wip/simdcomp-build 
Avoid using make for building simdcomp
 2017-03-21 10:00:49 +09:00
Laurentiu Nicola
92ce9b906b Avoid using make for building simdcomp 2017-03-21 00:25:04 +02:00
Laurentiu Nicola
1e0ac31e11 Clarify comment and use qualified import for the flag 2017-03-20 23:12:48 +02:00
Paul Masurel
ebcea0128c Getting the FLAG from the winapi module. 2017-03-19 11:09:15 +09:00
Paul Masurel
30075176cb blop 2017-03-19 10:52:54 +09:00
Laurentiu Nicola
7c114b602d Make directory syncing work on Windows 2017-03-19 02:17:13 +02:00
Paul Masurel
50659147d1 NOBUG updated simple_search.html 2017-03-14 12:04:21 +09:00
Paul Masurel
da10fe3b4d Various fixes. 2017-03-13 22:01:55 +09:00
Paul Masurel
4db56c6bd8 Merge pull request #101 from tantivy-search/issue/99 
Improvements to simple_search.rs: fixes #100 and improves #99
 2017-03-13 13:26:39 +09:00
Claus Matzinger
292dd6dcb6 fixup 2017-03-13 00:24:54 -04:00
Claus Matzinger
37e71f7c63 fixes #100 and improves #99 2017-03-12 22:59:38 -04:00
Paul Masurel
5932278e00 test passing 2017-03-13 10:00:19 +09:00
Paul Masurel
202dda98ba baby step 3 2017-03-12 19:00:57 +09:00
Paul Masurel
7c971b5d3b baby step 2 2017-03-11 16:14:20 +09:00
Paul Masurel
77c61ddab2 Baby step1 2017-03-11 14:20:46 +09:00
Paul Masurel
b7f026bab9 Merger returns a SegmentMeta 2017-03-10 09:05:51 +09:00
Paul Masurel
cc2f78184b Added unit test for #96 2017-03-10 09:05:51 +09:00
Paul Masurel
673423f762 Merge pull request #98 from KodrAus/feat/no-cpp 
Convert simd wrapper to C
 2017-03-09 13:11:08 +09:00
Paul Masurel
7532c4a440 Removed double ; 2017-03-09 10:57:30 +09:00
Ashley Mannix
324b56a60c fix warnings 2017-03-09 06:54:48 +10:00
Paul Masurel
ac3890f93c NOBUG Marked the functional test as ignore 2017-03-08 19:08:29 +09:00
Ashley Mannix
69b3de43f6 convert simd wrapper to c 2017-03-08 14:02:48 +10:00
Paul Masurel
3d1196d53e NOBUG added doc link. 2017-03-07 10:14:00 +09:00
Paul Masurel
a397537ed8 NOBUG added rustdoc 2017-03-07 10:10:43 +09:00
Paul Masurel
ebca904767 NOBUG added rustdoc 2017-03-07 09:58:51 +09:00
Paul Masurel
3a472914ce Fix .write -> .write_all 2017-03-06 16:28:30 +09:00
Paul Masurel
c59507444f issue/77 ManagedDirectory working 
Closes #77
 2017-03-06 12:18:36 +09:00
Paul Masurel
4b7afa2ae7 issue/77 Added managed directory 2017-03-03 22:41:37 +09:00
Paul Masurel
590a8582c9 The reference doc should not point to the schema page. 2017-02-28 21:17:19 +09:00
Paul Masurel
ab3440f925 NOBUG Bypass github cache for coveralls badge 2017-02-27 12:39:59 +09:00
Paul Masurel
ec5fb2eaa9 NOBUG cleanup 2017-02-27 09:52:28 +09:00
Paul Masurel
15b60d72cc NOBUG add_document does not return result 2017-02-27 09:36:41 +09:00
Paul Masurel
7a07144c68 Bugfix related with deletes, rollback and the index opstamp. 2017-02-27 01:42:25 +09:00
Paul Masurel
8bcfdb8e80 NOBUG misc ... 2017-02-26 21:35:18 +09:00
Paul Masurel
a7f10f055d Nobug hidding doc, filling doc 2017-02-26 00:11:32 +09:00
Paul Masurel
597dac9cb6 NOBUG Adding doc. 2017-02-25 23:39:02 +09:00
Paul Masurel
6a002bcc76 NOBUGwq 2017-02-25 21:20:55 +09:00
Paul Masurel
3a86fc00a2 Closes #64 - Improve Index creationg API / documentation 2017-02-25 20:40:39 +09:00
Paul Masurel
ca1617d3cd Fixes #91 2017-02-25 20:32:26 +09:00
Paul Masurel
e4a102d859 Merge branch 'issue/43' 
Conflicts:
	src/directory/mmap_directory.rs
 2017-02-25 19:36:21 +09:00
Paul Masurel
1d9924ee90 Closes #43. 2017-02-25 19:32:36 +09:00
Paul Masurel
f326a2dafe TODO hunt 2017-02-25 15:28:56 +09:00
Paul Masurel
78228ece73 Closes #92. ByteOrder of u32 terms. 2017-02-24 23:41:46 +09:00
Paul Masurel
503d0295cb issue/43 TODO hunt 2017-02-23 09:54:54 +09:00
Paul Masurel
eb39db44fc issue/43 Avoid keeping segments with 0 documents. 2017-02-23 09:20:30 +09:00
Paul Masurel
7f78d1f4ca Fixes #82 Renamed and commented the function to create Term from &[u8] 2017-02-23 08:33:59 +09:00
Paul Masurel
df9090cb0b NOBUG TODO hunt, and cleanups 2017-02-22 22:18:33 +09:00
Paul Masurel
4a8eb3cb05 issue/43 Added unit test for deletes including merging. 2017-02-22 21:38:37 +09:00
Paul Masurel
a74b41d7ed NOBUG run benchmark over exactly 100 K elements 2017-02-21 11:43:55 +09:00
Paul Masurel
06017bd422 NOBUG made the cleanup limit adaptive in MmapCache 2017-02-21 00:37:45 +09:00
Paul Masurel
17beaab8bf Merge branch 'issue/72' 2017-02-21 00:25:24 +09:00
Paul Masurel
062e38a2ab Fixes #72 - Cache directory uses weak ref. Introduced CacheInfo object. 2017-02-21 00:24:33 +09:00
Paul Masurel
8c2b20c496 NOBUG Trying to fix coverall conf. 2017-02-20 17:47:16 +09:00
Paul Masurel
c677eb9f13 issue/43 Removed notify 2017-02-19 22:41:45 +09:00
Paul Masurel
0f332d1fd3 issue/43 Removed doc freq from recorders. 2017-02-19 22:39:31 +09:00
Paul Masurel
1b45539f32 issue/43 Added support for delete in merged index 2017-02-19 22:39:31 +09:00
Paul Masurel
7315000fd4 issue/43 Merging ok for postings / fastfields. 2017-02-19 22:39:31 +09:00
Paul Masurel
e3d2fca844 issue/43 Isolated segment_entry / doc_opstamp_mapping 2017-02-19 22:39:31 +09:00
Paul Masurel
1c03d98a11 issue/43 added delete_queue right in the segment updater 2017-02-19 22:39:31 +09:00
Paul Masurel
8b68f22be1 issue/43 made the delete queue shareable 2017-02-19 22:39:31 +09:00
Paul Masurel
d007cf3435 issue/43 simplification. removed the notion of delete cursor. 2017-02-19 22:39:04 +09:00
Paul Masurel
72afbb28c7 issue/43 test passing 2017-02-19 22:39:04 +09:00
Paul Masurel
2fc3a505bc issue/43 refactoring segment meta 2017-02-19 22:39:04 +09:00
Paul Masurel
e337c35721 issue/43 SegmentMeta refactoring 2017-02-19 22:39:04 +09:00
Paul Masurel
0c318339b0 issue/43 Path logic in segment. 2017-02-19 22:39:04 +09:00
Paul Masurel
64fee11bc0 issue/43 Clean up 2017-02-19 22:39:04 +09:00
Paul Masurel
e12fc4bb09 issue/43 deletes 
merge not working
only updating uncommitted
 2017-02-19 22:39:04 +09:00
Paul Masurel
0820992141 issue/43 docstamp -> opstamp 2017-02-19 22:38:39 +09:00
Paul Masurel
09782858da issue/43 Segment have a commit opstamp 2017-02-19 22:38:39 +09:00
Paul Masurel
ca977fb17b issue/43 Refactoring of SegmentUpdater 2017-02-19 22:38:39 +09:00
Paul Masurel
e8ecb68f00 issue/43 switching for futures 2017-02-19 22:38:39 +09:00
Paul Masurel
0ec492dcf2 issue/43 refactoring in order to remove the segment updater non sense for simpler futures 2017-02-19 22:38:39 +09:00
Paul Masurel
20eb586660 issue/43 Rename SegmentUpdater 2017-02-19 22:38:39 +09:00
Paul Masurel
6530d43d6a issue/43 Small fixes. 2017-02-19 22:38:39 +09:00
Paul Masurel
926e71a573 issue/43 unit test running. segment updater uses futures. 2017-02-19 22:38:38 +09:00
Paul Masurel
bacaabf857 issue/43 fixed on unit test. need big refactoring of segment updater 2017-02-19 22:38:38 +09:00
Paul Masurel
d6e7157173 issue/43 Test broken... moved segment manager to the segment updater / segment writer 2017-02-19 22:38:15 +09:00
Paul Masurel
093dcbd253 issue/43 Isolated SegmentMeta 2017-02-19 22:38:15 +09:00
Paul Masurel
fba44b78b6 issue/43 Added delete doc file 2017-02-19 22:38:15 +09:00
Paul Masurel
01cf303dec issue/43 segment writer 2017-02-19 22:38:14 +09:00
Paul Masurel
d5c161e196 issue/43 Computing deleted doc bitset 2017-02-19 22:38:14 +09:00
Paul Masurel
183d5221b5 issue/43 DeleteQueue. 2017-02-19 22:38:14 +09:00
Paul Masurel
5a06f45403 issue/43 small progress 2017-02-19 22:36:57 +09:00
Paul Masurel
395cbf3913 issue/43 Change the delete queue datastruct for something cleaner/functional 2017-02-19 22:36:57 +09:00
Paul Masurel
fe2ddb8844 issue43 Added DeleteQueue. 2017-02-19 22:36:57 +09:00
Paul Masurel
3129701e92 issue/71 Added list of supported OSes 2017-02-19 14:14:15 +09:00
Paul Masurel
56ba698def Merge pull request #76 from Ameobea/master 
Updated dependency versions and implementations
 2017-02-17 18:20:44 +09:00
Casey Primozic
e0ba699c16 Updated dependency versions and implementations 
- Updated `byteorder` error usage (now returns straight `Error`s)
 - Updated `Uuid` implementation (`to_simple_string` now `.simple().to_string()`)
 2017-02-17 01:26:13 -06:00
Paul Masurel
b6423f9a76 Merge pull request #73 from manuel-woelker/pr-subtree 
Use git subtree mechanism for simdcomp to simplify build (cf. #24)
 2017-01-27 14:52:36 +09:00
Manuel Woelker
a667394a49 update README and build after simdcomp subtree refactor 2017-01-26 21:14:05 +01:00
Manuel Woelker
9f02b090dd Merge commit 'f07ccd6e4fbc5bbfeb94d40e0f14bc527a7d5439' as 'cpp/simdcomp' 2017-01-26 20:28:23 +01:00
Manuel Woelker
f07ccd6e4f Squashed 'cpp/simdcomp/' content from commit 0dca286 
git-subtree-dir: cpp/simdcomp
git-subtree-split: 0dca28668f1fb6d343dc3c62fa7750a00f1d7201
 2017-01-26 20:28:23 +01:00
Manuel Woelker
f19f8757de remove git submodule to replace via git subtree 2017-01-26 20:17:44 +01:00
Paul Masurel
f729edb529 NOBUG added badges / categories for crates.io 2017-01-21 09:35:44 +09:00
Paul Masurel
73ef201c44 Merge branch 'master' of github.com:tantivy-search/tantivy 2017-01-11 21:09:05 +09:00
Paul Masurel
3b69e790e9 NOBUG expose a version public api. Handy to check if the compilation was made with simd or not. 2017-01-11 21:06:41 +09:00
Paul Masurel
1b0b3051c2 NOBUG Pinned some version, removed import warning. 2017-01-09 15:30:50 +09:00
Paul Masurel
43c1da1a92 Merge branch 'issue/67' 2016-12-20 16:52:33 +01:00
Paul Masurel
e1cb5e299d NOBUG split field_type into 2 2016-12-20 16:51:34 +01:00
Paul Masurel
14ebed392b Merge pull request #68 from tantivy-search/issue/67 
Issue/67
 2016-12-20 11:27:05 +01:00
Paul Masurel
d3d34be167 issue/67 Added a advance interface to the term iterator 2016-12-20 11:25:52 +01:00
Paul Masurel
98cdc83428 Issue #67 Removing afterwards. 2016-12-18 11:57:28 +01:00
Paul Masurel
4d7d201f21 Issue #67 - Removed segment ord array from term iteration. 
This was probably an early optimization.
 2016-12-17 09:44:51 +01:00
Paul Masurel
ca5f3e1d46 issue/67 First stab. Iterator working. 2016-12-17 00:58:12 +01:00
Paul Masurel
1559733b03 Merge pull request #63 from vandenoever/readme 
fix for build instructions
 2016-12-12 10:17:27 +09:00
Paul Masurel
44b5f1868c Merge branch 'master' into readme 2016-12-12 10:17:19 +09:00
Paul Masurel
4cedfd903d NOBUG Added ga beacon to README 2016-12-12 10:07:30 +09:00
Paul Masurel
c0049e8487 NOBUG fixed doc urls. 2016-12-11 21:43:14 +09:00
Paul Masurel
e88adbff5c Bumped tantivy's version in Cargo.toml 2016-12-11 17:51:45 +09:00
Jos van den Oever
e497e04f70 fix for build instructions 
And clarification that nighty is required.
 2016-12-10 18:08:15 +01:00
208 changed files with 112480 additions and 6492 deletions
Expand all files Collapse all files

1
.gitattributes vendored Normal file
View File

@@ -0,0 +1 @@
cpp/* linguist-vendored

4
.gitignore vendored
View File

@@ -5,3 +5,7 @@ target/release
Cargo.lock
benchmark
.DS_Store
cpp/simdcomp/bitpackingbenchmark
*.bk
.idea
trace.dat

3
.gitmodules vendored
View File

@@ -1,3 +0,0 @@
[submodule "cpp/simdcomp"]
path = cpp/simdcomp
url = git@github.com:lemire/simdcomp.git

24
.travis.yml
View File

@@ -1,11 +1,7 @@
language: rust
cache: cargo
rust:
- nightly
git:
submodules: false
before_install:
- sed -i 's/git@github.com:/https:\/\/github.com\//' .gitmodules
- git submodule update --init --recursive
env:
global:
- CC=gcc-4.8
@@ -16,6 +12,7 @@ addons:
apt:
sources:
- ubuntu-toolchain-r-test
- kalakris-cmake
packages:
- gcc-4.8
- g++-4.8
@@ -23,18 +20,15 @@ addons:
- libelf-dev
- libdw-dev
- binutils-dev
- cmake
before_script:
- |
pip install 'travis-cargo<0.2' --user &&
export PATH=$HOME/.local/bin:$PATH
cargo install cargo-travis || echo "cargo-travis already installed"
export PATH=$HOME/.cargo/bin:$PATH
script:
- |
travis-cargo build &&
travis-cargo test &&
travis-cargo bench &&
travis-cargo doc
- cargo build
- cargo test
- cargo run --example simple_search
after_success:
- bash ./script/build-doc.sh
- cargo coveralls --exclude-pattern cpp/
- travis-cargo doc-upload
- if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then travis-cargo coveralls --no-sudo --verify; fi
- if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then ./kcov/build/src/kcov --verify --coveralls-id=$TRAVIS_JOB_ID --exclude-pattern=/.cargo target/kcov target/debug/tantivy-*; fi

100
CHANGELOG.md Normal file
View File

@@ -0,0 +1,100 @@
Tantivy 0.4.3
==========================
- Bugfix race condition when deleting files. (#198)
Tantivy 0.4.2
==========================
- Prevent usage of AVX2 instructions (#201)
Tantivy 0.4.1
==========================
- Bugfix for non-indexed fields. (#199)
Tantivy 0.4.0
==========================
- Raise the limit of number of fields (previously 256 fields) (@fulmicoton)
- Removed u32 fields. They are replaced by u64 and i64 fields (#65) (@fulmicoton)
- Optimized skip in SegmentPostings (#130) (@lnicola)
- Replacing rustc_serialize by serde. Kudos to @KodrAus and @lnicola
- Using error-chain (@KodrAus)
- QueryParser: (@fulmicoton)
- Explicit error returned when searched for a term that is not indexed
- Searching for a int term via the query parser was broken `(age:1)`
- Searching for a non-indexed field returns an explicit Error
- Phrase query for non-tokenized field are not tokenized by the query parser.
- Faster/Better indexing (@fulmicoton)
- using murmurhash2
- faster merging
- more memory efficient fast field writer (@lnicola )
- better handling of collisions
- lesser memory usage
- Added API, most notably to iterate over ranges of terms (@fulmicoton)
- Bugfix that was preventing to unmap segment files, on index drop (@fulmicoton)
- Made the doc! macro public (@fulmicoton)
- Added an alternative implementation of the streaming dictionary (@fulmicoton)
Tantivy 0.3.1
==========================
- Expose a method to trigger files garbage collection
Tantivy 0.3
==========================
Special thanks to @Kodraus @lnicola @Ameobea @manuel-woelker @celaus
for their contribution to this release.
Thanks also to everyone in tantivy gitter chat
for their advise and company :)
https://gitter.im/tantivy-search/tantivy
Warning:
Tantivy 0.3 is NOT backward compatible with tantivy 0.2
code and index format.
You should not expect backward compatibility before
tantivy 1.0.
New Features
------------
- Delete. You can now delete documents from an index.
- Support for windows (Thanks to @lnicola)
Various Bugfixes & small improvements
----------------------------------------
- Added CI for Windows (https://ci.appveyor.com/project/fulmicoton/tantivy)
Thanks to @KodrAus ! (#108)
- Various dependy version update (Thanks to @Ameobea) #76
- Fixed several race conditions in `Index.wait_merge_threads`
- Fixed #72. Mmap were never released.
- Fixed #80. Fast field used to take an amplitude of 32 bits after a merge. (Ouch!)
- Fixed #92. u32 are now encoded using big endian in the fst
in order to make there enumeration consistent with
the natural ordering.
- Building binary targets for tantivy-cli (Thanks to @KodrAus)
- Misc invisible bug fixes, and code cleanup.
- Use

65
Cargo.toml
View File

@@ -1,46 +1,58 @@
[package]
name = "tantivy"
version = "0.1.1"
version = "0.5.0-dev"
authors = ["Paul Masurel <paul.masurel@gmail.com>"]
build = "build.rs"
license = "MIT"
categories = ["database-implementations", "data-structures"]
description = """Tantivy is a search engine library."""
documentation = "http://fulmicoton.com/tantivy/tantivy/index.html"
documentation = "https://tantivy-search.github.io/tantivy/tantivy/index.html"
homepage = "https://github.com/tantivy-search/tantivy"
repository = "https://github.com/tantivy-search/tantivy"
readme = "README.md"
keywords = ["search", "information", "retrieval"]
[dependencies]
byteorder = "0.4"
memmap = "0.2"
lazy_static = "0.1"
regex = "0.1"
fst = "0.1"
atomicwrites = "0.0.14"
tempfile = "2.0"
rustc-serialize = "0.3"
log = "0.3"
combine = "2.0.*"
byteorder = "1.0"
lazy_static = "0.2.1"
tinysegmenter = "0.1.0"
regex = "0.2"
fst = "0.2"
atomicwrites = "0.1.3"
tempfile = "2.1"
log = "0.3.6"
combine = "2.2"
tempdir = "0.3"
bincode = "0.4"
libc = {version = "0.2.6", optional=true}
num_cpus = "0.2"
itertools = "0.4"
lz4 = "1.13"
serde = "1.0"
serde_derive = "1.0"
serde_json = "1.0"
bincode = "0.8"
libc = {version = "0.2.20", optional=true}
num_cpus = "1.2"
itertools = "0.5.9"
lz4 = "1.20"
bit-set = "0.4.0"
time = "0.1"
uuid = "0.1"
uuid = { version = "0.5", features = ["v4", "serde"] }
chan = "0.1"
crossbeam = "0.2"
version = "2"
crossbeam = "0.3"
futures = "0.1"
futures-cpupool = "0.1"
error-chain = "0.8"
owning_ref = "0.3"
stable_deref_trait = "1.0.0"
rust-stemmers = "0.1.0"
[target.'cfg(windows)'.dependencies]
winapi = "0.2"
[dev-dependencies]
rand = "0.3"
env_logger = "0.4"
[build-dependencies]
gcc = {version = "0.3", optional=true}
cc = {version = "1.0.0", optional=true}
[profile.release]
opt-level = 3
@@ -51,4 +63,9 @@ debug-assertions = false
[features]
default = ["simdcompression"]
simdcompression = ["libc", "gcc"]
simdcompression = ["libc", "cc"]
streamdict = []
[badges]
travis-ci = { repository = "tantivy-search/tantivy" }

41
README.md
View File

@@ -1,45 +1,50 @@
![Tantivy](https://tantivy-search.github.io/logo/tantivy-logo.png)
[![Build Status](https://travis-ci.org/tantivy-search/tantivy.svg?branch=master)](https://travis-ci.org/tantivy-search/tantivy)
[![Coverage Status](https://coveralls.io/repos/github/tantivy-search/tantivy/badge.svg?branch=master)](https://coveralls.io/github/tantivy-search/tantivy?branch=master)
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
[![Coverage Status](https://coveralls.io/repos/github/tantivy-search/tantivy/badge.svg?branch=master&refresh1)](https://coveralls.io/github/tantivy-search/tantivy?branch=master)
[![Join the chat at https://gitter.im/tantivy-search/tantivy](https://badges.gitter.im/tantivy-search/tantivy.svg)](https://gitter.im/tantivy-search/tantivy?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge&utm_content=badge)
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
[![Build status](https://ci.appveyor.com/api/projects/status/r7nb13kj23u8m9pj?svg=true)](https://ci.appveyor.com/project/fulmicoton/tantivy)
![beacon for google analytics](https://ga-beacon.appspot.com/UA-88834340-1/tantivy/README)
**Tantivy** is a **full text search engine library** written in rust.
It is strongly inspired by Lucene's design.
# Features
- configurable indexing (optional term frequency and position indexing)
- tf-idf scoring
- Basic query language
- Phrase queries
- Incremental indexing
- Multithreaded indexing (indexing English Wikipedia takes 4 minutes on my desktop)
- Multithreaded indexing (indexing English Wikipedia takes < 3 minutes on my desktop)
- mmap based
- SIMD integer compression
- u32 fast fields (equivalent of doc values in Lucene)
- optional SIMD integer compression
- u64 and i64 fast fields (equivalent of doc values in Lucene)
- LZ4 compressed document store
- Cheesy logo with a horse
Tantivy supports Linux, MacOS and Windows.
# Getting started
- [tantivy's usage example](http://fulmicoton.com/tantivy-examples/simple_search.html)
- [tantivy-cli and its tutorial](https://github.com/fulmicoton/tantivy-cli).
- [tantivy-cli and its tutorial](https://github.com/tantivy-search/tantivy-cli).
It will walk you through getting a wikipedia search engine up and running in a few minutes.
- [reference doc](http://fulmicoton.com/tantivy/tantivy/index.html).
- [reference doc]
- [For the last released version](https://docs.rs/tantivy/)
- [For the last master branch](https://tantivy-search.github.io/tantivy/tantivy/index.html)
# Compiling
# Compiling
Tantivy requires Rust Nightly because it uses requires the features [`box_syntax`](https://doc.rust-lang.org/stable/book/box-syntax-and-patterns.html), [`optin_builtin_traits`](https://github.com/rust-lang/rfcs/blob/master/text/0019-opt-in-builtin-traits.md), and [`conservative_impl_trait`](https://github.com/rust-lang/rfcs/blob/master/text/1522-conservative-impl-trait.md).
The project can then be built using `cargo`.
By default, `tantivy` uses a git submodule called `simdcomp`.
After cloning the repository, you will need to initialize and update
the submodules. The project can then be built using `cargo`.
git clone git@github.com:fulmicoton/tantivy.git
git submodule init
git submodule update
git clone git@github.com:tantivy-search/tantivy.git
cd tantivy
cargo build
@@ -47,9 +52,9 @@ Alternatively, if you are trying to compile `tantivy` without simd compression,
you can disable this functionality. In this case, this submodule is not required
and you can compile tantivy by using the `--no-default-features` flag.
cargo build --no-default-features
cargo build --no-default-features
# Contribute
Send me an email (paul.masurel at gmail.com) if you want to contribute to tantivy.
Send me an email (paul.masurel at gmail.com) if you want to contribute to tantivy.

25
appveyor.yml Normal file
View File

@@ -0,0 +1,25 @@
# Appveyor configuration template for Rust using rustup for Rust installation
# https://github.com/starkat99/appveyor-rust
os: Visual Studio 2015
environment:
matrix:
- channel: nightly
target: x86_64-pc-windows-msvc
- channel: nightly
target: x86_64-pc-windows-gnu
msys_bits: 64
install:
- appveyor DownloadFile https://win.rustup.rs/ -FileName rustup-init.exe
- rustup-init -yv --default-toolchain %channel% --default-host %target%
- set PATH=%PATH%;%USERPROFILE%\.cargo\bin
- if defined msys_bits set PATH=%PATH%;C:\msys64\mingw%msys_bits%\bin
- rustc -vV
- cargo -vV
build: false
test_script:
- REM SET RUST_LOG=tantivy,test & cargo test --verbose
- REM SET RUST_BACKTRACE=1 & cargo run --example simple_search

77
build.rs
View File

@@ -1,40 +1,61 @@
#[cfg(feature= "simdcompression")]
#[cfg(feature = "simdcompression")]
mod build {
extern crate gcc;
use std::process::Command;
extern crate cc;
pub fn build() {
Command::new("make")
.current_dir("cpp/simdcomp")
.output()
.unwrap_or_else(|e| { panic!("Failed to make simdcomp: {}", e) });
gcc::Config::new()
.cpp(true)
.flag("-std=c++11")
.flag("-O3")
.flag("-mssse3")
.include("./cpp/simdcomp/include")
.object("cpp/simdcomp/avxbitpacking.o")
.object("cpp/simdcomp/simdintegratedbitpacking.o")
.object("cpp/simdcomp/simdbitpacking.o")
.object("cpp/simdcomp/simdpackedsearch.o")
.object("cpp/simdcomp/simdcomputil.o")
.object("cpp/simdcomp/simdpackedselect.o")
.object("cpp/simdcomp/simdfor.o")
.file("cpp/simdcomp_wrapper.cpp")
.compile("libsimdcomp.a");
println!("cargo:rustc-flags=-l dylib=stdc++");
let mut config = cc::Build::new();
config
.include("./cpp/simdcomp/include")
.file("cpp/simdcomp/src/avxbitpacking.c")
.file("cpp/simdcomp/src/simdintegratedbitpacking.c")
.file("cpp/simdcomp/src/simdbitpacking.c")
.file("cpp/simdcomp/src/simdpackedsearch.c")
.file("cpp/simdcomp/src/simdcomputil.c")
.file("cpp/simdcomp/src/simdpackedselect.c")
.file("cpp/simdcomp/src/simdfor.c")
.file("cpp/simdcomp_wrapper.c");
if !cfg!(debug_assertions) {
config.opt_level(3);
if cfg!(target_env = "msvc") {
config
.define("NDEBUG", None)
.flag("/Gm-")
.flag("/GS-")
.flag("/Gy")
.flag("/Oi")
.flag("/GL");
}
}
if !cfg!(target_env = "msvc") {
config
.include("./cpp/streamvbyte/include")
.file("cpp/streamvbyte/src/streamvbyte.c")
.file("cpp/streamvbyte/src/streamvbytedelta.c")
.flag("-msse4.1")
.flag("-march=native")
.flag("-std=c99");
}
config.compile("libsimdcomp.a");
// Workaround for linking static libraries built with /GL
// https://github.com/rust-lang/rust/issues/26003
if !cfg!(debug_assertions) && cfg!(target_env = "msvc") {
println!("cargo:rustc-link-lib=dylib=simdcomp");
}
println!("cargo:rerun-if-changed=cpp");
}
}
#[cfg(not(feature= "simdcompression"))]
#[cfg(not(feature = "simdcompression"))]
mod build {
pub fn build() {
}
pub fn build() {}
}
fn main() {
build::build();
}

1
cpp/simdcomp vendored

Submodule cpp/simdcomp deleted from 0dca28668f

9
cpp/simdcomp/.gitignore vendored Normal file
View File

@@ -0,0 +1,9 @@
Makefile.in
lib*
unit*
*.o
src/*.lo
src/*.o
src/.deps
src/.dirstamp
src/.libs

11
cpp/simdcomp/.travis.yml Normal file
View File

@@ -0,0 +1,11 @@
language: c
sudo: false
compiler:
- gcc
- clang
branches:
only:
- master
script: make && ./unit

9
cpp/simdcomp/CHANGELOG Normal file
View File

@@ -0,0 +1,9 @@
Upcoming
- added missing include
- improved portability (MSVC)
- implemented C89 compatibility
Version 0.0.3 (19 May 2014)
- improved documentation
Version 0.0.2 (6 February 2014)
- added go demo
Version 0.0.1 (5 February 2014)

27
cpp/simdcomp/LICENSE Normal file
View File

@@ -0,0 +1,27 @@
Copyright (c) 2014--, The authors
All rights reserved.
Redistribution and use in source and binary forms, with or without modification,
are permitted provided that the following conditions are met:
* Redistributions of source code must retain the above copyright notice, this
list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright notice, this
list of conditions and the following disclaimer in the documentation and/or
other materials provided with the distribution.
* Neither the name of the {organization} nor the names of its
contributors may be used to endorse or promote products derived from
this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

137
cpp/simdcomp/README.md Normal file
View File

@@ -0,0 +1,137 @@
The SIMDComp library
====================
[![Build Status](https://travis-ci.org/lemire/simdcomp.png)](https://travis-ci.org/lemire/simdcomp)
A simple C library for compressing lists of integers using binary packing and SIMD instructions.
The assumption is either that you have a list of 32-bit integers where most of them are small, or a list of 32-bit integers where differences between successive integers are small. No software is able to reliably compress an array of 32-bit random numbers.
This library can decode at least 4 billions of compressed integers per second on most
desktop or laptop processors. That is, it can decompress data at a rate of 15 GB/s.
This is significantly faster than generic codecs like gzip, LZO, Snappy or LZ4.
On a Skylake Intel processor, it can decode integers at a rate 0.3 cycles per integer,
which can easily translate into more than 8 decoded billions integers per second.
Contributors: Daniel Lemire, Nathan Kurz, Christoph Rupp, Anatol Belski, Nick White and others
What is it for?
-------------
This is a low-level library for fast integer compression. By design it does not define a compressed
format. It is up to the (sophisticated) user to create a compressed format.
Requirements
-------------
- Your processor should support SSE4.1 (It is supported by most Intel and AMD processors released since 2008.)
- It is possible to build the core part of the code if your processor support SSE2 (Pentium4 or better)
- C99 compliant compiler (GCC is assumed)
- A Linux-like distribution is assumed by the makefile
For a plain C version that does not use SIMD instructions, see https://github.com/lemire/LittleIntPacker
Usage
-------
Compression works over blocks of 128 integers.
For a complete working example, see example.c (you can build it and
run it with "make example; ./example").
1) Lists of integers in random order.
```C
const uint32_t b = maxbits(datain);// computes bit width
simdpackwithoutmask(datain, buffer, b);//compressed to buffer, compressing 128 32-bit integers down to b*32 bytes
simdunpack(buffer, backbuffer, b);//uncompressed to backbuffer
```
While 128 32-bit integers are read, only b 128-bit words are written. Thus, the compression ratio is 32/b.
2) Sorted lists of integers.
We used differential coding: we store the difference between successive integers. For this purpose, we need an initial value (called offset).
```C
uint32_t offset = 0;
uint32_t b1 = simdmaxbitsd1(offset,datain); // bit width
simdpackwithoutmaskd1(offset, datain, buffer, b1);//compressing 128 32-bit integers down to b1*32 bytes
simdunpackd1(offset, buffer, backbuffer, b1);//uncompressed
```
General example for arrays of arbitrary length:
```C
int compress_decompress_demo() {
size_t k, N = 9999;
__m128i * endofbuf;
uint32_t * datain = malloc(N * sizeof(uint32_t));
uint8_t * buffer;
uint32_t * backbuffer = malloc(N * sizeof(uint32_t));
uint32_t b;
for (k = 0; k < N; ++k){ /* start with k=0, not k=1! */
datain[k] = k;
}
b = maxbits_length(datain, N);
buffer = malloc(simdpack_compressedbytes(N,b)); // allocate just enough memory
endofbuf = simdpack_length(datain, N, (__m128i *)buffer, b);
/* compressed data is stored between buffer and endofbuf using (endofbuf-buffer)*sizeof(__m128i) bytes */
/* would be safe to do : buffer = realloc(buffer,(endofbuf-(__m128i *)buffer)*sizeof(__m128i)); */
simdunpack_length((const __m128i *)buffer, N, backbuffer, b);
for (k = 0; k < N; ++k){
if(datain[k] != backbuffer[k]) {
printf("bug\n");
return -1;
}
}
return 0;
}
```
3) Frame-of-Reference
We also have frame-of-reference (FOR) functions (see simdfor.h header). They work like the bit packing
routines, but do not use differential coding so they allow faster search in some cases, at the expense
of compression.
Setup
---------
make
make test
and if you are daring:
make install
Go
--------
If you are a go user, there is a "go" folder where you will find a simple demo.
Other libraries
----------------
* Fast decoder for VByte-compressed integers https://github.com/lemire/MaskedVByte
* Fast integer compression in C using StreamVByte https://github.com/lemire/streamvbyte
* FastPFOR is a C++ research library well suited to compress unsorted arrays: https://github.com/lemire/FastPFor
* SIMDCompressionAndIntersection is a C++ research library well suited for sorted arrays (differential coding)
and computing intersections: https://github.com/lemire/SIMDCompressionAndIntersection
* TurboPFor is a C library that offers lots of interesting optimizations. Well worth checking! (GPL license) https://github.com/powturbo/TurboPFor
* Oroch is a C++ library that offers a usable API (MIT license) https://github.com/ademakov/Oroch
References
------------
* Daniel Lemire, Leonid Boytsov, Nathan Kurz, SIMD Compression and the Intersection of Sorted Integers, Software Practice & Experience 46 (6) 2016. http://arxiv.org/abs/1401.6399
* Daniel Lemire and Leonid Boytsov, Decoding billions of integers per second through vectorization, Software Practice & Experience 45 (1), 2015. http://arxiv.org/abs/1209.2137 http://onlinelibrary.wiley.com/doi/10.1002/spe.2203/abstract
* Jeff Plaisance, Nathan Kurz, Daniel Lemire, Vectorized VByte Decoding, International Symposium on Web Algorithms 2015, 2015. http://arxiv.org/abs/1503.07387
* Wayne Xin Zhao, Xudong Zhang, Daniel Lemire, Dongdong Shan, Jian-Yun Nie, Hongfei Yan, Ji-Rong Wen, A General SIMD-based Approach to Accelerating Compression Algorithms, ACM Transactions on Information Systems 33 (3), 2015. http://arxiv.org/abs/1502.01916
* T. D. Wu, Bitpacking techniques for indexing genomes: I. Hash tables, Algorithms for Molecular Biology 11 (5), 2016. http://almob.biomedcentral.com/articles/10.1186/s13015-016-0069-5

235
cpp/simdcomp/benchmarks/benchmark.c Normal file
View File

@@ -0,0 +1,235 @@
/**
* This code is released under a BSD License.
*/
#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include "simdcomp.h"
#ifdef _MSC_VER
# include <windows.h>
__int64 freq;
typedef __int64 time_snap_t;
static time_snap_t time_snap(void)
{
__int64 now;
QueryPerformanceCounter((LARGE_INTEGER *)&now);
return (__int64)((now*1000000)/freq);
}
# define TIME_SNAP_FMT "%I64d"
#else
# define time_snap clock
# define TIME_SNAP_FMT "%lu"
typedef clock_t time_snap_t;
#endif
void benchmarkSelect() {
uint32_t buffer[128];
uint32_t backbuffer[128];
uint32_t initial = 33;
uint32_t b;
time_snap_t S1, S2, S3;
int i;
printf("benchmarking select \n");
/* this test creates delta encoded buffers with different bits, then
* performs lower bound searches for each key */
for (b = 0; b <= 32; b++) {
uint32_t prev = initial;
uint32_t out[128];
/* initialize the buffer */
for (i = 0; i < 128; i++) {
buffer[i] = ((uint32_t)(1655765 * i )) ;
if(b < 32) buffer[i] %= (1<<b);
}
for (i = 0; i < 128; i++) {
buffer[i] = buffer[i] + prev;
prev = buffer[i];
}
for (i = 1; i < 128; i++) {
if(buffer[i] < buffer[i-1] )
buffer[i] = buffer[i-1];
}
assert(simdmaxbitsd1(initial, buffer)<=b);
for (i = 0; i < 128; i++) {
out[i] = 0; /* memset would do too */
}
/* delta-encode to 'i' bits */
simdpackwithoutmaskd1(initial, buffer, (__m128i *)out, b);
S1 = time_snap();
for (i = 0; i < 128 * 10; i++) {
uint32_t valretrieved = simdselectd1(initial, (__m128i *)out, b, (uint32_t)i % 128);
assert(valretrieved == buffer[i%128]);
}
S2 = time_snap();
for (i = 0; i < 128 * 10; i++) {
simdunpackd1(initial, (__m128i *)out, backbuffer, b);
assert(backbuffer[i % 128] == buffer[i % 128]);
}
S3 = time_snap();
printf("bit width = %d, fast select function time = " TIME_SNAP_FMT ", naive time = " TIME_SNAP_FMT " \n", b, (S2-S1), (S3-S2));
}
}
int uint32_cmp(const void *a, const void *b)
{
const uint32_t *ia = (const uint32_t *)a;
const uint32_t *ib = (const uint32_t *)b;
if(*ia < *ib)
return -1;
else if (*ia > *ib)
return 1;
return 0;
}
/* adapted from wikipedia */
int binary_search(uint32_t * A, uint32_t key, int imin, int imax)
{
int imid;
imax --;
while(imin + 1 < imax) {
imid = imin + ((imax - imin) / 2);
if (A[imid] > key) {
imax = imid;
} else if (A[imid] < key) {
imin = imid;
} else {
return imid;
}
}
return imax;
}
/* adapted from wikipedia */
int lower_bound(uint32_t * A, uint32_t key, int imin, int imax)
{
int imid;
imax --;
while(imin + 1 < imax) {
imid = imin + ((imax - imin) / 2);
if (A[imid] >= key) {
imax = imid;
} else if (A[imid] < key) {
imin = imid;
}
}
if(A[imin] >= key) return imin;
return imax;
}
void benchmarkSearch() {
uint32_t buffer[128];
uint32_t backbuffer[128];
uint32_t out[128];
uint32_t result, initial = 0;
uint32_t b, i;
time_snap_t S1, S2, S3, S4;
printf("benchmarking search \n");
/* this test creates delta encoded buffers with different bits, then
* performs lower bound searches for each key */
for (b = 0; b <= 32; b++) {
uint32_t prev = initial;
/* initialize the buffer */
for (i = 0; i < 128; i++) {
buffer[i] = ((uint32_t)rand()) ;
if(b < 32) buffer[i] %= (1<<b);
}
qsort(buffer,128, sizeof(uint32_t), uint32_cmp);
for (i = 0; i < 128; i++) {
buffer[i] = buffer[i] + prev;
prev = buffer[i];
}
for (i = 1; i < 128; i++) {
if(buffer[i] < buffer[i-1] )
buffer[i] = buffer[i-1];
}
assert(simdmaxbitsd1(initial, buffer)<=b);
for (i = 0; i < 128; i++) {
out[i] = 0; /* memset would do too */
}
/* delta-encode to 'i' bits */
simdpackwithoutmaskd1(initial, buffer, (__m128i *)out, b);
simdunpackd1(initial, (__m128i *)out, backbuffer, b);
for (i = 0; i < 128; i++) {
assert(buffer[i] == backbuffer[i]);
}
S1 = time_snap();
for (i = 0; i < 128 * 10; i++) {
int pos;
uint32_t pseudorandomkey = buffer[i%128];
__m128i vecinitial = _mm_set1_epi32(initial);
pos = simdsearchd1(&vecinitial, (__m128i *)out, b,
pseudorandomkey, &result);
if((result < pseudorandomkey) || (buffer[pos] != result)) {
printf("bug A.\n");
} else if (pos > 0) {
if(buffer[pos-1] >= pseudorandomkey)
printf("bug B.\n");
}
}
S2 = time_snap();
for (i = 0; i < 128 * 10; i++) {
int pos;
uint32_t pseudorandomkey = buffer[i%128];
simdunpackd1(initial, (__m128i *)out, backbuffer, b);
pos = lower_bound(backbuffer, pseudorandomkey, 0, 128);
result = backbuffer[pos];
if((result < pseudorandomkey) || (buffer[pos] != result)) {
printf("bug C.\n");
} else if (pos > 0) {
if(buffer[pos-1] >= pseudorandomkey)
printf("bug D.\n");
}
}
S3 = time_snap();
for (i = 0; i < 128 * 10; i++) {
int pos;
uint32_t pseudorandomkey = buffer[i%128];
pos = simdsearchwithlengthd1(initial, (__m128i *)out, b, 128,
pseudorandomkey, &result);
if((result < pseudorandomkey) || (buffer[pos] != result)) {
printf("bug A.\n");
} else if (pos > 0) {
if(buffer[pos-1] >= pseudorandomkey)
printf("bug B.\n");
}
}
S4 = time_snap();
printf("bit width = %d, fast search function time = " TIME_SNAP_FMT ", naive time = " TIME_SNAP_FMT " , fast with length time = " TIME_SNAP_FMT " \n", b, (S2-S1), (S3-S2), (S4-S3) );
}
}
int main() {
#ifdef _MSC_VER
QueryPerformanceFrequency((LARGE_INTEGER *)&freq);
#endif
benchmarkSearch();
benchmarkSelect();
return 0;
}

205
cpp/simdcomp/benchmarks/bitpackingbenchmark.c Normal file
View File

@@ -0,0 +1,205 @@
#include <stdio.h>
#include "simdcomp.h"
#define RDTSC_START(cycles) \
do { \
register unsigned cyc_high, cyc_low; \
__asm volatile( \
"cpuid\n\t" \
"rdtsc\n\t" \
"mov %%edx, %0\n\t" \
"mov %%eax, %1\n\t" \
: "=r"(cyc_high), "=r"(cyc_low)::"%rax", "%rbx", "%rcx", "%rdx"); \
(cycles) = ((uint64_t)cyc_high << 32) | cyc_low; \
} while (0)
#define RDTSC_FINAL(cycles) \
do { \
register unsigned cyc_high, cyc_low; \
__asm volatile( \
"rdtscp\n\t" \
"mov %%edx, %0\n\t" \
"mov %%eax, %1\n\t" \
"cpuid\n\t" \
: "=r"(cyc_high), "=r"(cyc_low)::"%rax", "%rbx", "%rcx", "%rdx"); \
(cycles) = ((uint64_t)cyc_high << 32) | cyc_low; \
} while (0)
uint32_t * get_random_array_from_bit_width(uint32_t length, uint32_t bit) {
uint32_t * answer = malloc(sizeof(uint32_t) * length);
uint32_t mask = (uint32_t) ((UINT64_C(1) << bit) - 1);
uint32_t i;
for(i = 0; i < length; ++i) {
answer[i] = rand() & mask;
}
return answer;
}
uint32_t * get_random_array_from_bit_width_d1(uint32_t length, uint32_t bit) {
uint32_t * answer = malloc(sizeof(uint32_t) * length);
uint32_t mask = (uint32_t) ((UINT64_C(1) << bit) - 1);
uint32_t i;
answer[0] = rand() & mask;
for(i = 1; i < length; ++i) {
answer[i] = answer[i-1] + (rand() & mask);
}
return answer;
}
void demo128() {
const uint32_t length = 128;
uint32_t bit;
printf("# --- %s\n", __func__);
printf("# compressing %d integers\n",length);
printf("# format: bit width, pack in cycles per int, unpack in cycles per int\n");
for(bit = 1; bit <= 32; ++bit) {
uint32_t i;
uint32_t * data = get_random_array_from_bit_width(length, bit);
__m128i * buffer = malloc(length * sizeof(uint32_t));
uint32_t * backdata = malloc(length * sizeof(uint32_t));
uint32_t repeat = 500;
uint64_t min_diff;
printf("%d\t",bit);
min_diff = (uint64_t)-1;
for (i = 0; i < repeat; i++) {
uint64_t cycles_start, cycles_final, cycles_diff;
__asm volatile("" ::: /* pretend to clobber */ "memory");
RDTSC_START(cycles_start);
simdpackwithoutmask(data,buffer, bit);
RDTSC_FINAL(cycles_final);
cycles_diff = (cycles_final - cycles_start);
if (cycles_diff < min_diff) min_diff = cycles_diff;
}
printf("%.2f\t",min_diff*1.0/length);
min_diff = (uint64_t)-1;
for (i = 0; i < repeat; i++) {
uint64_t cycles_start, cycles_final, cycles_diff;
__asm volatile("" ::: /* pretend to clobber */ "memory");
RDTSC_START(cycles_start);
simdunpack(buffer, backdata,bit);
RDTSC_FINAL(cycles_final);
cycles_diff = (cycles_final - cycles_start);
if (cycles_diff < min_diff) min_diff = cycles_diff;
}
printf("%.2f\t",min_diff*1.0/length);
free(data);
free(buffer);
free(backdata);
printf("\n");
}
printf("\n\n"); /* two blank lines are required by gnuplot */
}
void demo128_d1() {
const uint32_t length = 128;
uint32_t bit;
printf("# --- %s\n", __func__);
printf("# compressing %d integers\n",length);
printf("# format: bit width, pack in cycles per int, unpack in cycles per int\n");
for(bit = 1; bit <= 32; ++bit) {
uint32_t i;
uint32_t * data = get_random_array_from_bit_width_d1(length, bit);
__m128i * buffer = malloc(length * sizeof(uint32_t));
uint32_t * backdata = malloc(length * sizeof(uint32_t));
uint32_t repeat = 500;
uint64_t min_diff;
printf("%d\t",bit);
min_diff = (uint64_t)-1;
for (i = 0; i < repeat; i++) {
uint64_t cycles_start, cycles_final, cycles_diff;
__asm volatile("" ::: /* pretend to clobber */ "memory");
RDTSC_START(cycles_start);
simdpackwithoutmaskd1(0,data,buffer, bit);
RDTSC_FINAL(cycles_final);
cycles_diff = (cycles_final - cycles_start);
if (cycles_diff < min_diff) min_diff = cycles_diff;
}
printf("%.2f\t",min_diff*1.0/length);
min_diff = (uint64_t)-1;
for (i = 0; i < repeat; i++) {
uint64_t cycles_start, cycles_final, cycles_diff;
__asm volatile("" ::: /* pretend to clobber */ "memory");
RDTSC_START(cycles_start);
simdunpackd1(0,buffer, backdata,bit);
RDTSC_FINAL(cycles_final);
cycles_diff = (cycles_final - cycles_start);
if (cycles_diff < min_diff) min_diff = cycles_diff;
}
printf("%.2f\t",min_diff*1.0/length);
free(data);
free(buffer);
free(backdata);
printf("\n");
}
printf("\n\n"); /* two blank lines are required by gnuplot */
}
#ifdef __AVX2__
void demo256() {
const uint32_t length = 256;
uint32_t bit;
printf("# --- %s\n", __func__);
printf("# compressing %d integers\n",length);
printf("# format: bit width, pack in cycles per int, unpack in cycles per int\n");
for(bit = 1; bit <= 32; ++bit) {
uint32_t i;
uint32_t * data = get_random_array_from_bit_width(length, bit);
__m256i * buffer = malloc(length * sizeof(uint32_t));
uint32_t * backdata = malloc(length * sizeof(uint32_t));
uint32_t repeat = 500;
uint64_t min_diff;
printf("%d\t",bit);
min_diff = (uint64_t)-1;
for (i = 0; i < repeat; i++) {
uint64_t cycles_start, cycles_final, cycles_diff;
__asm volatile("" ::: /* pretend to clobber */ "memory");
RDTSC_START(cycles_start);
avxpackwithoutmask(data,buffer, bit);
RDTSC_FINAL(cycles_final);
cycles_diff = (cycles_final - cycles_start);
if (cycles_diff < min_diff) min_diff = cycles_diff;
}
printf("%.2f\t",min_diff*1.0/length);
min_diff = (uint64_t)-1;
for (i = 0; i < repeat; i++) {
uint64_t cycles_start, cycles_final, cycles_diff;
__asm volatile("" ::: /* pretend to clobber */ "memory");
RDTSC_START(cycles_start);
avxunpack(buffer, backdata,bit);
RDTSC_FINAL(cycles_final);
cycles_diff = (cycles_final - cycles_start);
if (cycles_diff < min_diff) min_diff = cycles_diff;
}
printf("%.2f\t",min_diff*1.0/length);
free(data);
free(buffer);
free(backdata);
printf("\n");
}
printf("\n\n"); /* two blank lines are required by gnuplot */
}
#endif /* avx 2 */
int main() {
demo128();
demo128_d1();
#ifdef __AVX2__
demo256();
#endif
return 0;
}

195
cpp/simdcomp/example.c Normal file
View File

@@ -0,0 +1,195 @@
/* Type "make example" to build this example program. */
#include <stdio.h>
#include <time.h>
#include <stdlib.h>
#include "simdcomp.h"
/**
We provide several different code examples.
**/
/* very simple test to illustrate a simple application */
int compress_decompress_demo() {
size_t k, N = 9999;
__m128i * endofbuf;
int howmanybytes;
float compratio;
uint32_t * datain = malloc(N * sizeof(uint32_t));
uint8_t * buffer;
uint32_t * backbuffer = malloc(N * sizeof(uint32_t));
uint32_t b;
printf("== simple test\n");
for (k = 0; k < N; ++k) { /* start with k=0, not k=1! */
datain[k] = k;
}
b = maxbits_length(datain, N);
buffer = malloc(simdpack_compressedbytes(N,b));
endofbuf = simdpack_length(datain, N, (__m128i *)buffer, b);
howmanybytes = (endofbuf-(__m128i *)buffer)*sizeof(__m128i); /* number of compressed bytes */
compratio = N*sizeof(uint32_t) * 1.0 / howmanybytes;
/* endofbuf points to the end of the compressed data */
buffer = realloc(buffer,(endofbuf-(__m128i *)buffer)*sizeof(__m128i)); /* optional but safe. */
printf("Compressed %d integers down to %d bytes (comp. ratio = %f).\n",(int)N,howmanybytes,compratio);
/* in actual applications b must be stored and retrieved: caller is responsible for that. */
simdunpack_length((const __m128i *)buffer, N, backbuffer, b); /* will return a pointer to endofbuf */
for (k = 0; k < N; ++k) {
if(datain[k] != backbuffer[k]) {
printf("bug at %lu \n",(unsigned long)k);
return -1;
}
}
printf("Code works!\n");
free(datain);
free(buffer);
free(backbuffer);
return 0;
}
/* compresses data from datain to buffer, returns how many bytes written
used below in simple_demo */
size_t compress(uint32_t * datain, size_t length, uint8_t * buffer) {
uint32_t offset;
uint8_t * initout;
size_t k;
if(length/SIMDBlockSize*SIMDBlockSize != length) {
printf("Data length should be a multiple of %i \n",SIMDBlockSize);
}
offset = 0;
initout = buffer;
for(k = 0; k < length / SIMDBlockSize; ++k) {
uint32_t b = simdmaxbitsd1(offset,
datain + k * SIMDBlockSize);
*buffer++ = b;
simdpackwithoutmaskd1(offset, datain + k * SIMDBlockSize, (__m128i *) buffer,
b);
offset = datain[k * SIMDBlockSize + SIMDBlockSize - 1];
buffer += b * sizeof(__m128i);
}
return buffer - initout;
}
/* Another illustration ... */
void simple_demo() {
size_t REPEAT = 10, gap;
size_t N = 1000 * SIMDBlockSize;/* SIMDBlockSize is 128 */
uint32_t * datain = malloc(N * sizeof(uint32_t));
size_t compsize;
clock_t start, end;
uint8_t * buffer = malloc(N * sizeof(uint32_t) + N / SIMDBlockSize); /* output buffer */
uint32_t * backbuffer = malloc(SIMDBlockSize * sizeof(uint32_t));
printf("== simple demo\n");
for (gap = 1; gap <= 243; gap *= 3) {
size_t k, repeat;
uint32_t offset = 0;
uint32_t bogus = 0;
double numberofseconds;
printf("\n");
printf(" gap = %lu \n", (unsigned long) gap);
datain[0] = 0;
for (k = 1; k < N; ++k)
datain[k] = datain[k-1] + ( rand() % (gap + 1) );
compsize = compress(datain,N,buffer);
printf("compression ratio = %f \n", (N * sizeof(uint32_t))/ (compsize * 1.0 ));
start = clock();
for(repeat = 0; repeat < REPEAT; ++repeat) {
uint8_t * decbuffer = buffer;
for (k = 0; k * SIMDBlockSize < N; ++k) {
uint8_t b = *decbuffer++;
simdunpackd1(offset, (__m128i *) decbuffer, backbuffer, b);
/* do something here with backbuffer */
bogus += backbuffer[3];
decbuffer += b * sizeof(__m128i);
offset = backbuffer[SIMDBlockSize - 1];
}
}
end = clock();
numberofseconds = (end-start)/(double)CLOCKS_PER_SEC;
printf("decoding speed in million of integers per second %f \n",N*REPEAT/(numberofseconds*1000.0*1000.0));
start = clock();
for(repeat = 0; repeat < REPEAT; ++repeat) {
uint8_t * decbuffer = buffer;
for (k = 0; k * SIMDBlockSize < N; ++k) {
memcpy(backbuffer,decbuffer+k*SIMDBlockSize,SIMDBlockSize*sizeof(uint32_t));
bogus += backbuffer[3] - backbuffer[100];
}
}
end = clock();
numberofseconds = (end-start)/(double)CLOCKS_PER_SEC;
printf("memcpy speed in million of integers per second %f \n",N*REPEAT/(numberofseconds*1000.0*1000.0));
printf("ignore me %i \n",bogus);
printf("All tests are in CPU cache. Avoid out-of-cache decoding in applications.\n");
}
free(buffer);
free(datain);
free(backbuffer);
}
/* Used below in more_sophisticated_demo ... */
size_t varying_bit_width_compress(uint32_t * datain, size_t length, uint8_t * buffer) {
uint8_t * initout;
size_t k;
if(length/SIMDBlockSize*SIMDBlockSize != length) {
printf("Data length should be a multiple of %i \n",SIMDBlockSize);
}
initout = buffer;
for(k = 0; k < length / SIMDBlockSize; ++k) {
uint32_t b = maxbits(datain);
*buffer++ = b;
simdpackwithoutmask(datain, (__m128i *)buffer, b);
datain += SIMDBlockSize;
buffer += b * sizeof(__m128i);
}
return buffer - initout;
}
/* Here we compress the data in blocks of 128 integers with varying bit width */
int varying_bit_width_demo() {
size_t nn = 128 * 2;
uint32_t * datainn = malloc(nn * sizeof(uint32_t));
uint8_t * buffern = malloc(nn * sizeof(uint32_t) + nn / SIMDBlockSize);
uint8_t * initbuffern = buffern;
uint32_t * backbuffern = malloc(nn * sizeof(uint32_t));
size_t k, compsize;
printf("== varying bit-width demo\n");
for(k=0; k<nn; ++k) {
datainn[k] = rand() % (k + 1);
}
compsize = varying_bit_width_compress(datainn,nn,buffern);
printf("encoded size: %u (original size: %u)\n", (unsigned)compsize,
(unsigned)(nn * sizeof(uint32_t)));
for (k = 0; k * SIMDBlockSize < nn; ++k) {
uint32_t b = *buffern;
buffern++;
simdunpack((const __m128i *)buffern, backbuffern + k * SIMDBlockSize, b);
buffern += b * sizeof(__m128i);
}
for (k = 0; k < nn; ++k) {
if(backbuffern[k] != datainn[k]) {
printf("bug\n");
return -1;
}
}
printf("Code works!\n");
free(datainn);
free(initbuffern);
free(backbuffern);
return 0;
}
int main() {
if(compress_decompress_demo() != 0) return -1;
if(varying_bit_width_demo() != 0) return -1;
simple_demo();
return 0;
}

13
cpp/simdcomp/go/README.md Normal file
View File

@@ -0,0 +1,13 @@
Simple Go demo
==============
Setup
======
Start by installing the simdcomp library (make && make install).
Then type:
go run test.go

71
cpp/simdcomp/go/test.go Normal file
View File

@@ -0,0 +1,71 @@
/////////
// This particular file is in the public domain.
// Author: Daniel Lemire
////////
package main
/*
#cgo LDFLAGS: -lsimdcomp
#include <simdcomp.h>
*/
import "C"
import "fmt"
//////////
// For this demo, we pack and unpack blocks of 128 integers
/////////
func main() {
// I am going to use C types. Alternative might be to use unsafe.Pointer calls, see http://bit.ly/1ndw3W3
// this is our original data
var data [128]C.uint32_t
for i := C.uint32_t(0); i < C.uint32_t(128); i++ {
data[i] = i
}
////////////
// We first pack without differential coding
///////////
// computing how many bits per int. is needed
b := C.maxbits(&data[0])
ratio := 32.0/float64(b)
fmt.Println("Bit width ", b)
fmt.Println(fmt.Sprintf("Compression ratio %f ", ratio))
// we are now going to create a buffer to receive the packed data (each __m128i uses 128 bits)
out := make([] C.__m128i,b)
C.simdpackwithoutmask( &data[0],&out[0],b);
var recovereddata [128]C.uint32_t
C.simdunpack(&out[0],&recovereddata[0],b)
for i := 0; i < 128; i++ {
if data[i] != recovereddata[i] {
fmt.Println("Bug ")
return
}
}
///////////
// Next, we use differential coding
//////////
offset := C.uint32_t(0) // if you pack data from K to K + 128, offset should be the value at K-1. When K = 0, choose a default
b1 := C.simdmaxbitsd1(offset,&data[0])
ratio1 := 32.0/float64(b1)
fmt.Println("Bit width ", b1)
fmt.Println(fmt.Sprintf("Compression ratio %f ", ratio1))
// we are now going to create a buffer to receive the packed data (each __m128i uses 128 bits)
out = make([] C.__m128i,b1)
C.simdpackwithoutmaskd1(offset, &data[0],&out[0],b1);
C.simdunpackd1(offset,&out[0],&recovereddata[0],b1)
for i := 0; i < 128; i++ {
if data[i] != recovereddata[i] {
fmt.Println("Bug ")
return
}
}
fmt.Println("test succesful.")
}

40
cpp/simdcomp/include/avxbitpacking.h Normal file
View File

@@ -0,0 +1,40 @@
/**
* This code is released under a BSD License.
*/
#ifndef INCLUDE_AVXBITPACKING_H_
#define INCLUDE_AVXBITPACKING_H_
#ifdef __AVX2__
#include "portability.h"
/* AVX2 is required */
#include <immintrin.h>
/* for memset */
#include <string.h>
#include "simdcomputil.h"
enum{ AVXBlockSize = 256};
/* max integer logarithm over a range of AVXBlockSize integers (256 integer) */
uint32_t avxmaxbits(const uint32_t * begin);
/* reads 256 values from "in", writes "bit" 256-bit vectors to "out" */
void avxpack(const uint32_t * in,__m256i * out, const uint32_t bit);
/* reads 256 values from "in", writes "bit" 256-bit vectors to "out" */
void avxpackwithoutmask(const uint32_t * in,__m256i * out, const uint32_t bit);
/* reads "bit" 256-bit vectors from "in", writes 256 values to "out" */
void avxunpack(const __m256i * in,uint32_t * out, const uint32_t bit);
#endif /* __AVX2__ */
#endif /* INCLUDE_AVXBITPACKING_H_ */

81
cpp/simdcomp/include/portability.h Normal file
View File

@@ -0,0 +1,81 @@
/**
* This code is released under a BSD License.
*/
#ifndef SIMDBITCOMPAT_H_
#define SIMDBITCOMPAT_H_
#include <iso646.h> /* mostly for Microsoft compilers */
#include <string.h>
#if SIMDCOMP_DEBUG
# define SIMDCOMP_ALWAYS_INLINE inline
# define SIMDCOMP_NEVER_INLINE
# define SIMDCOMP_PURE
#else
# if defined(__GNUC__)
# if __GNUC__ >= 3
# define SIMDCOMP_ALWAYS_INLINE inline __attribute__((always_inline))
# define SIMDCOMP_NEVER_INLINE __attribute__((noinline))
# define SIMDCOMP_PURE __attribute__((pure))
# else
# define SIMDCOMP_ALWAYS_INLINE inline
# define SIMDCOMP_NEVER_INLINE
# define SIMDCOMP_PURE
# endif
# elif defined(_MSC_VER)
# define SIMDCOMP_ALWAYS_INLINE __forceinline
# define SIMDCOMP_NEVER_INLINE
# define SIMDCOMP_PURE
# else
# if __has_attribute(always_inline)
# define SIMDCOMP_ALWAYS_INLINE inline __attribute__((always_inline))
# else
# define SIMDCOMP_ALWAYS_INLINE inline
# endif
# if __has_attribute(noinline)
# define SIMDCOMP_NEVER_INLINE __attribute__((noinline))
# else
# define SIMDCOMP_NEVER_INLINE
# endif
# if __has_attribute(pure)
# define SIMDCOMP_PURE __attribute__((pure))
# else
# define SIMDCOMP_PURE
# endif
# endif
#endif
#if defined(_MSC_VER) && _MSC_VER < 1600
typedef unsigned int uint32_t;
typedef unsigned char uint8_t;
typedef signed char int8_t;
#else
#include <stdint.h> /* part of Visual Studio 2010 and better, others likely anyway */
#endif
#if defined(_MSC_VER)
#define SIMDCOMP_ALIGNED(x) __declspec(align(x))
#else
#if defined(__GNUC__)
#define SIMDCOMP_ALIGNED(x) __attribute__ ((aligned(x)))
#endif
#endif
#if defined(_MSC_VER)
# include <intrin.h>
/* 64-bit needs extending */
# define SIMDCOMP_CTZ(result, mask) do { \
unsigned long index; \
if (!_BitScanForward(&(index), (mask))) { \
(result) = 32U; \
} else { \
(result) = (uint32_t)(index); \
} \
} while (0)
#else
# define SIMDCOMP_CTZ(result, mask) \
result = __builtin_ctz(mask)
#endif
#endif /* SIMDBITCOMPAT_H_ */

72
cpp/simdcomp/include/simdbitpacking.h Normal file
View File

@@ -0,0 +1,72 @@
/**
* This code is released under a BSD License.
*/
#ifndef SIMDBITPACKING_H_
#define SIMDBITPACKING_H_
#include "portability.h"
/* SSE2 is required */
#include <emmintrin.h>
/* for memset */
#include <string.h>
#include "simdcomputil.h"
/***
* Please see example.c for various examples on how to make good use
* of these functions.
*/
/* reads 128 values from "in", writes "bit" 128-bit vectors to "out".
* The input values are masked so that only the least significant "bit" bits are used. */
void simdpack(const uint32_t * in,__m128i * out, const uint32_t bit);
/* reads 128 values from "in", writes "bit" 128-bit vectors to "out".
* The input values are assumed to be less than 1<<bit. */
void simdpackwithoutmask(const uint32_t * in,__m128i * out, const uint32_t bit);
/* reads "bit" 128-bit vectors from "in", writes 128 values to "out" */
void simdunpack(const __m128i * in,uint32_t * out, const uint32_t bit);
/* how many compressed bytes are needed to compressed length integers using a bit width of bit with
the simdpackFOR_length function. */
int simdpack_compressedbytes(int length, const uint32_t bit);
/* like simdpack, but supports an undetermined number of inputs.
* This is useful if you need to unpack an array of integers that is not divisible by 128 integers.
* Returns a pointer to the (advanced) compressed array. Compressed data is stored in the memory location between
the provided (out) pointer and the returned pointer. */
__m128i * simdpack_length(const uint32_t * in, size_t length, __m128i * out, const uint32_t bit);
/* like simdunpack, but supports an undetermined number of inputs.
* This is useful if you need to unpack an array of integers that is not divisible by 128 integers.
* Returns a pointer to the (advanced) compressed array. The read compressed data is between the provided
(in) pointer and the returned pointer. */
const __m128i * simdunpack_length(const __m128i * in, size_t length, uint32_t * out, const uint32_t bit);
/* like simdpack, but supports an undetermined small number of inputs. This is useful if you need to pack less
than 128 integers.
* Note that this function is much slower.
* Returns a pointer to the (advanced) compressed array. Compressed data is stored in the memory location
between the provided (out) pointer and the returned pointer. */
__m128i * simdpack_shortlength(const uint32_t * in, int length, __m128i * out, const uint32_t bit);
/* like simdunpack, but supports an undetermined small number of inputs. This is useful if you need to unpack less
than 128 integers.
* Note that this function is much slower.
* Returns a pointer to the (advanced) compressed array. The read compressed data is between the provided (in)
pointer and the returned pointer. */
const __m128i * simdunpack_shortlength(const __m128i * in, int length, uint32_t * out, const uint32_t bit);
/* given a block of 128 packed values, this function sets the value at index "index" to "value" */
void simdfastset(__m128i * in128, uint32_t b, uint32_t value, size_t index);
#endif /* SIMDBITPACKING_H_ */

22
cpp/simdcomp/include/simdcomp.h Normal file
View File

@@ -0,0 +1,22 @@
/**
* This code is released under a BSD License.
*/
#ifndef SIMDCOMP_H_
#define SIMDCOMP_H_
#ifdef __cplusplus
extern "C" {
#endif
#include "simdbitpacking.h"
#include "simdcomputil.h"
#include "simdfor.h"
#include "simdintegratedbitpacking.h"
#include "avxbitpacking.h"
#ifdef __cplusplus
} // extern "C"
#endif
#endif

54
cpp/simdcomp/include/simdcomputil.h Normal file
View File

@@ -0,0 +1,54 @@
/**
* This code is released under a BSD License.
*/
#ifndef SIMDCOMPUTIL_H_
#define SIMDCOMPUTIL_H_
#include "portability.h"
/* SSE2 is required */
#include <emmintrin.h>
/* returns the integer logarithm of v (bit width) */
uint32_t bits(const uint32_t v);
/* max integer logarithm over a range of SIMDBlockSize integers (128 integer) */
uint32_t maxbits(const uint32_t * begin);
/* same as maxbits, but we specify the number of integers */
uint32_t maxbits_length(const uint32_t * in,uint32_t length);
enum{ SIMDBlockSize = 128};
/* computes (quickly) the minimal value of 128 values */
uint32_t simdmin(const uint32_t * in);
/* computes (quickly) the minimal value of the specified number of values */
uint32_t simdmin_length(const uint32_t * in, uint32_t length);
#ifdef __SSE4_1__
/* computes (quickly) the minimal and maximal value of the specified number of values */
void simdmaxmin_length(const uint32_t * in, uint32_t length, uint32_t * getmin, uint32_t * getmax);
/* computes (quickly) the minimal and maximal value of the 128 values */
void simdmaxmin(const uint32_t * in, uint32_t * getmin, uint32_t * getmax);
#endif
/* like maxbit over 128 integers (SIMDBlockSize) with provided initial value
and using differential coding */
uint32_t simdmaxbitsd1(uint32_t initvalue, const uint32_t * in);
/* like simdmaxbitsd1, but calculates maxbits over |length| integers
with provided initial value. |length| can be any arbitrary value. */
uint32_t simdmaxbitsd1_length(uint32_t initvalue, const uint32_t * in,
uint32_t length);
#endif /* SIMDCOMPUTIL_H_ */

72
cpp/simdcomp/include/simdfor.h Normal file
View File

@@ -0,0 +1,72 @@
/**
* This code is released under a BSD License.
*/
#ifndef INCLUDE_SIMDFOR_H_
#define INCLUDE_SIMDFOR_H_
#include "portability.h"
/* SSE2 is required */
#include <emmintrin.h>
#include "simdcomputil.h"
#include "simdbitpacking.h"
#ifdef __cplusplus
extern "C" {
#endif
/* reads 128 values from "in", writes "bit" 128-bit vectors to "out" */
void simdpackFOR(uint32_t initvalue, const uint32_t * in,__m128i * out, const uint32_t bit);
/* reads "bit" 128-bit vectors from "in", writes 128 values to "out" */
void simdunpackFOR(uint32_t initvalue, const __m128i * in,uint32_t * out, const uint32_t bit);
/* how many compressed bytes are needed to compressed length integers using a bit width of bit with
the simdpackFOR_length function. */
int simdpackFOR_compressedbytes(int length, const uint32_t bit);
/* like simdpackFOR, but supports an undetermined number of inputs.
This is useful if you need to pack less than 128 integers. Note that this function is much slower.
Compressed data is stored in the memory location between
the provided (out) pointer and the returned pointer. */
__m128i * simdpackFOR_length(uint32_t initvalue, const uint32_t * in, int length, __m128i * out, const uint32_t bit);
/* like simdunpackFOR, but supports an undetermined number of inputs.
This is useful if you need to unpack less than 128 integers. Note that this function is much slower.
The read compressed data is between the provided
(in) pointer and the returned pointer. */
const __m128i * simdunpackFOR_length(uint32_t initvalue, const __m128i * in, int length, uint32_t * out, const uint32_t bit);
/* returns the value stored at the specified "slot".
* */
uint32_t simdselectFOR(uint32_t initvalue, const __m128i *in, uint32_t bit,
int slot);
/* given a block of 128 packed values, this function sets the value at index "index" to "value" */
void simdfastsetFOR(uint32_t initvalue, __m128i * in, uint32_t bit, uint32_t value, size_t index);
/* searches "bit" 128-bit vectors from "in" (= length<=128 encoded integers) for the first encoded uint32 value
* which is >= |key|, and returns its position. It is assumed that the values
* stored are in sorted order.
* The encoded key is stored in "*presult".
* The first length decoded integers, ignoring others. If no value is larger or equal to the key,
* length is returned. Length should be no larger than 128.
*
* If no value is larger or equal to the key,
* length is returned */
int simdsearchwithlengthFOR(uint32_t initvalue, const __m128i *in, uint32_t bit,
int length, uint32_t key, uint32_t *presult);
#ifdef __cplusplus
} // extern "C"
#endif
#endif /* INCLUDE_SIMDFOR_H_ */

98
cpp/simdcomp/include/simdintegratedbitpacking.h Normal file
View File

@@ -0,0 +1,98 @@
/**
* This code is released under a BSD License.
*/
#ifndef SIMD_INTEGRATED_BITPACKING_H
#define SIMD_INTEGRATED_BITPACKING_H
#include "portability.h"
/* SSE2 is required */
#include <emmintrin.h>
#include "simdcomputil.h"
#include "simdbitpacking.h"
#ifdef __cplusplus
extern "C" {
#endif
/* reads 128 values from "in", writes "bit" 128-bit vectors to "out"
integer values should be in sorted order (for best results).
The differences are masked so that only the least significant "bit" bits are used. */
void simdpackd1(uint32_t initvalue, const uint32_t * in,__m128i * out, const uint32_t bit);
/* reads 128 values from "in", writes "bit" 128-bit vectors to "out"
integer values should be in sorted order (for best results).
The difference values are assumed to be less than 1<<bit. */
void simdpackwithoutmaskd1(uint32_t initvalue, const uint32_t * in,__m128i * out, const uint32_t bit);
/* reads "bit" 128-bit vectors from "in", writes 128 values to "out" */
void simdunpackd1(uint32_t initvalue, const __m128i * in,uint32_t * out, const uint32_t bit);
/* searches "bit" 128-bit vectors from "in" (= 128 encoded integers) for the first encoded uint32 value
* which is >= |key|, and returns its position. It is assumed that the values
* stored are in sorted order.
* The encoded key is stored in "*presult". If no value is larger or equal to the key,
* 128 is returned. The pointer initOffset is a pointer to the last four value decoded
* (when starting out, this can be a zero vector or initialized with _mm_set1_epi32(init)),
* and the vector gets updated.
**/
int
simdsearchd1(__m128i * initOffset, const __m128i *in, uint32_t bit,
uint32_t key, uint32_t *presult);
/* searches "bit" 128-bit vectors from "in" (= length<=128 encoded integers) for the first encoded uint32 value
* which is >= |key|, and returns its position. It is assumed that the values
* stored are in sorted order.
* The encoded key is stored in "*presult".
* The first length decoded integers, ignoring others. If no value is larger or equal to the key,
* length is returned. Length should be no larger than 128.
*
* If no value is larger or equal to the key,
* length is returned */
int simdsearchwithlengthd1(uint32_t initvalue, const __m128i *in, uint32_t bit,
int length, uint32_t key, uint32_t *presult);
/* returns the value stored at the specified "slot".
* */
uint32_t simdselectd1(uint32_t initvalue, const __m128i *in, uint32_t bit,
int slot);
/* given a block of 128 packed values, this function sets the value at index "index" to "value",
* you must somehow know the previous value.
* Because of differential coding, all following values are incremented by the offset between this new
* value and the old value...
* This functions is useful if you want to modify the last value.
*/
void simdfastsetd1fromprevious( __m128i * in, uint32_t bit, uint32_t previousvalue, uint32_t value, size_t index);
/* given a block of 128 packed values, this function sets the value at index "index" to "value",
* This function computes the previous value if needed.
* Because of differential coding, all following values are incremented by the offset between this new
* value and the old value...
* This functions is useful if you want to modify the last value.
*/
void simdfastsetd1(uint32_t initvalue, __m128i * in, uint32_t bit, uint32_t value, size_t index);
/*Simply scan the data
* The pointer initOffset is a pointer to the last four value decoded
* (when starting out, this can be a zero vector or initialized with _mm_set1_epi32(init);),
* and the vector gets updated.
* */
void
simdscand1(__m128i * initOffset, const __m128i *in, uint32_t bit);
#ifdef __cplusplus
} // extern "C"
#endif
#endif

79
cpp/simdcomp/makefile Normal file
View File

@@ -0,0 +1,79 @@
# minimalist makefile
.SUFFIXES:
#
.SUFFIXES: .cpp .o .c .h
ifeq ($(DEBUG),1)
CFLAGS = -fPIC -std=c89 -ggdb -msse4.1 -march=native -Wall -Wextra -Wshadow -fsanitize=undefined -fno-omit-frame-pointer -fsanitize=address
else
CFLAGS = -fPIC -std=c89 -O3 -msse4.1 -march=native -Wall -Wextra -Wshadow
endif # debug
LDFLAGS = -shared
LIBNAME=libsimdcomp.so.0.0.3
all: unit unit_chars bitpackingbenchmark $(LIBNAME)
test:
./unit
./unit_chars
install: $(OBJECTS)
cp $(LIBNAME) /usr/local/lib
ln -s /usr/local/lib/$(LIBNAME) /usr/local/lib/libsimdcomp.so
ldconfig
cp $(HEADERS) /usr/local/include
HEADERS=./include/simdbitpacking.h ./include/simdcomputil.h ./include/simdintegratedbitpacking.h ./include/simdcomp.h ./include/simdfor.h ./include/avxbitpacking.h
uninstall:
for h in $(HEADERS) ; do rm /usr/local/$$h; done
rm /usr/local/lib/$(LIBNAME)
rm /usr/local/lib/libsimdcomp.so
ldconfig
OBJECTS= simdbitpacking.o simdintegratedbitpacking.o simdcomputil.o \
simdpackedsearch.o simdpackedselect.o simdfor.o avxbitpacking.o
$(LIBNAME): $(OBJECTS)
$(CC) $(CFLAGS) -o $(LIBNAME) $(OBJECTS) $(LDFLAGS)
avxbitpacking.o: ./src/avxbitpacking.c $(HEADERS)
$(CC) $(CFLAGS) -c ./src/avxbitpacking.c -Iinclude
simdfor.o: ./src/simdfor.c $(HEADERS)
$(CC) $(CFLAGS) -c ./src/simdfor.c -Iinclude
simdcomputil.o: ./src/simdcomputil.c $(HEADERS)
$(CC) $(CFLAGS) -c ./src/simdcomputil.c -Iinclude
simdbitpacking.o: ./src/simdbitpacking.c $(HEADERS)
$(CC) $(CFLAGS) -c ./src/simdbitpacking.c -Iinclude
simdintegratedbitpacking.o: ./src/simdintegratedbitpacking.c $(HEADERS)
$(CC) $(CFLAGS) -c ./src/simdintegratedbitpacking.c -Iinclude
simdpackedsearch.o: ./src/simdpackedsearch.c $(HEADERS)
$(CC) $(CFLAGS) -c ./src/simdpackedsearch.c -Iinclude
simdpackedselect.o: ./src/simdpackedselect.c $(HEADERS)
$(CC) $(CFLAGS) -c ./src/simdpackedselect.c -Iinclude
example: ./example.c $(HEADERS) $(OBJECTS)
$(CC) $(CFLAGS) -o example ./example.c -Iinclude $(OBJECTS)
unit: ./tests/unit.c $(HEADERS) $(OBJECTS)
$(CC) $(CFLAGS) -o unit ./tests/unit.c -Iinclude $(OBJECTS)
bitpackingbenchmark: ./benchmarks/bitpackingbenchmark.c $(HEADERS) $(OBJECTS)
$(CC) $(CFLAGS) -o bitpackingbenchmark ./benchmarks/bitpackingbenchmark.c -Iinclude $(OBJECTS)
benchmark: ./benchmarks/benchmark.c $(HEADERS) $(OBJECTS)
$(CC) $(CFLAGS) -o benchmark ./benchmarks/benchmark.c -Iinclude $(OBJECTS)
dynunit: ./tests/unit.c $(HEADERS) $(LIBNAME)
$(CC) $(CFLAGS) -o dynunit ./tests/unit.c -Iinclude -lsimdcomp
unit_chars: ./tests/unit_chars.c $(HEADERS) $(OBJECTS)
$(CC) $(CFLAGS) -o unit_chars ./tests/unit_chars.c -Iinclude $(OBJECTS)
clean:
rm -f unit *.o $(LIBNAME) example benchmark bitpackingbenchmark dynunit unit_chars

104
cpp/simdcomp/makefile.vc Normal file
View File

@@ -0,0 +1,104 @@
!IFNDEF MACHINE
!IF "$(PROCESSOR_ARCHITECTURE)"=="AMD64"
MACHINE=x64
!ELSE
MACHINE=x86
!ENDIF
!ENDIF
!IFNDEF DEBUG
DEBUG=no
!ENDIF
!IFNDEF CC
CC=cl.exe
!ENDIF
!IFNDEF AR
AR=lib.exe
!ENDIF
!IFNDEF LINK
LINK=link.exe
!ENDIF
!IFNDEF PGO
PGO=no
!ENDIF
!IFNDEF PGI
PGI=no
!ENDIF
INC = /Iinclude
!IF "$(DEBUG)"=="yes"
CFLAGS = /nologo /MDd /LDd /Od /Zi /D_DEBUG /RTC1 /W3 /GS /Gm
ARFLAGS = /nologo
LDFLAGS = /nologo /debug /nodefaultlib:msvcrt
!ELSE
CFLAGS = /nologo /MD /O2 /Zi /DNDEBUG /W3 /Gm- /GS /Gy /Oi /GL /MP
ARFLAGS = /nologo /LTCG
LDFLAGS = /nologo /LTCG /DYNAMICBASE /incremental:no /debug /opt:ref,icf
!ENDIF
!IF "$(PGI)"=="yes"
LDFLAGS = $(LDFLAGS) /ltcg:pgi
!ENDIF
!IF "$(PGO)"=="yes"
LDFLAGS = $(LDFLAGS) /ltcg:pgo
!ENDIF
LIB_OBJS = simdbitpacking.obj simdintegratedbitpacking.obj simdcomputil.obj \
simdpackedsearch.obj simdpackedselect.obj simdfor.obj
all: lib dll dynunit unit_chars example benchmark
# need some good use case scenario to train the instrumented build
@if "$(PGI)"=="yes" echo Running PGO training
@if "$(PGI)"=="yes" benchmark.exe >nul 2>&1
@if "$(PGI)"=="yes" example.exe >nul 2>&1
$(LIB_OBJS):
$(CC) $(INC) $(CFLAGS) /c src/simdbitpacking.c src/simdintegratedbitpacking.c src/simdcomputil.c \
src/simdpackedsearch.c src/simdpackedselect.c src/simdfor.c
lib: $(LIB_OBJS)
$(AR) $(ARFLAGS) /OUT:simdcomp_a.lib $(LIB_OBJS)
dll: $(LIB_OBJS)
$(LINK) /DLL $(LDFLAGS) /OUT:simdcomp.dll /IMPLIB:simdcomp.lib /DEF:simdcomp.def $(LIB_OBJS)
unit: lib
$(CC) $(INC) $(CFLAGS) /c src/unit.c
$(LINK) $(LDFLAGS) /OUT:unit.exe unit.obj simdcomp_a.lib
dynunit: dll
$(CC) $(INC) $(CFLAGS) /c src/unit.c
$(LINK) $(LDFLAGS) /OUT:unit.exe unit.obj simdcomp.lib
unit_chars: lib
$(CC) $(INC) $(CFLAGS) /c src/unit_chars.c
$(LINK) $(LDFLAGS) /OUT:unit_chars.exe unit_chars.obj simdcomp.lib
example: lib
$(CC) $(INC) $(CFLAGS) /c example.c
$(LINK) $(LDFLAGS) /OUT:example.exe example.obj simdcomp.lib
benchmark: lib
$(CC) $(INC) $(CFLAGS) /c src/benchmark.c
$(LINK) $(LDFLAGS) /OUT:benchmark.exe benchmark.obj simdcomp.lib
clean:
del /Q *.obj
del /Q *.lib
del /Q *.exe
del /Q *.dll
del /Q *.pgc
del /Q *.pgd
del /Q *.pdb

16
cpp/simdcomp/package.json Normal file
View File

@@ -0,0 +1,16 @@
{
"name": "simdcomp",
"version": "0.0.3",
"repo": "lemire/simdcomp",
"description": "A simple C library for compressing lists of integers",
"license": "BSD-3-Clause",
"src": [
"src/simdbitpacking.c",
"src/simdcomputil.c",
"src/simdintegratedbitpacking.c",
"include/simdbitpacking.h",
"include/simdcomp.h",
"include/simdcomputil.h",
"include/simdintegratedbitpacking.h"
]
}

182
cpp/simdcomp/scripts/avxpacking.py Executable file
View File

@@ -0,0 +1,182 @@
#!/usr/bin/env python
import sys
def howmany(bit):
""" how many values are we going to pack? """
return 256
def howmanywords(bit):
return (howmany(bit) * bit + 255)/256
def howmanybytes(bit):
return howmanywords(bit) * 16
print("""
/** code generated by avxpacking.py starts here **/
""")
print("""typedef void (*avxpackblockfnc)(const uint32_t * pin, __m256i * compressed);""")
print("""typedef void (*avxunpackblockfnc)(const __m256i * compressed, uint32_t * pout);""")
def plurial(number):
if(number <> 1):
return "s"
else :
return ""
print("")
print("static void avxpackblock0(const uint32_t * pin, __m256i * compressed) {");
print(" (void)compressed;");
print(" (void) pin; /* we consumed {0} 32-bit integer{1} */ ".format(howmany(0),plurial(howmany(0))));
print("}");
print("")
for bit in range(1,33):
print("")
print("/* we are going to pack {0} {1}-bit values, touching {2} 256-bit words, using {3} bytes */ ".format(howmany(bit),bit,howmanywords(bit),howmanybytes(bit)))
print("static void avxpackblock{0}(const uint32_t * pin, __m256i * compressed) {{".format(bit));
print(" const __m256i * in = (const __m256i *) pin;");
print(" /* we are going to touch {0} 256-bit word{1} */ ".format(howmanywords(bit),plurial(howmanywords(bit))));
if(howmanywords(bit) == 1):
print(" __m256i w0;")
else:
print(" __m256i w0, w1;")
if( (bit & (bit-1)) <> 0) : print(" __m256i tmp; /* used to store inputs at word boundary */")
oldword = 0
for j in range(howmany(bit)/8):
firstword = j * bit / 32
if(firstword > oldword):
print(" _mm256_storeu_si256(compressed + {0}, w{1});".format(oldword,oldword%2))
oldword = firstword
secondword = (j * bit + bit - 1)/32
firstshift = (j*bit) % 32
if( firstword == secondword):
if(firstshift == 0):
print(" w{0} = _mm256_lddqu_si256 (in + {1});".format(firstword%2,j))
else:
print(" w{0} = _mm256_or_si256(w{0},_mm256_slli_epi32(_mm256_lddqu_si256 (in + {1}) , {2}));".format(firstword%2,j,firstshift))
else:
print(" tmp = _mm256_lddqu_si256 (in + {0});".format(j))
print(" w{0} = _mm256_or_si256(w{0},_mm256_slli_epi32(tmp , {2}));".format(firstword%2,j,firstshift))
secondshift = 32-firstshift
print(" w{0} = _mm256_srli_epi32(tmp,{2});".format(secondword%2,j,secondshift))
print(" _mm256_storeu_si256(compressed + {0}, w{1});".format(secondword,secondword%2))
print("}");
print("")
print("")
print("static void avxpackblockmask0(const uint32_t * pin, __m256i * compressed) {");
print(" (void)compressed;");
print(" (void) pin; /* we consumed {0} 32-bit integer{1} */ ".format(howmany(0),plurial(howmany(0))));
print("}");
print("")
for bit in range(1,33):
print("")
print("/* we are going to pack {0} {1}-bit values, touching {2} 256-bit words, using {3} bytes */ ".format(howmany(bit),bit,howmanywords(bit),howmanybytes(bit)))
print("static void avxpackblockmask{0}(const uint32_t * pin, __m256i * compressed) {{".format(bit));
print(" /* we are going to touch {0} 256-bit word{1} */ ".format(howmanywords(bit),plurial(howmanywords(bit))));
if(howmanywords(bit) == 1):
print(" __m256i w0;")
else:
print(" __m256i w0, w1;")
print(" const __m256i * in = (const __m256i *) pin;");
if(bit < 32): print(" const __m256i mask = _mm256_set1_epi32({0});".format((1<<bit)-1));
def maskfnc(x):
if(bit == 32): return x
return " _mm256_and_si256 ( mask, {0}) ".format(x)
if( (bit & (bit-1)) <> 0) : print(" __m256i tmp; /* used to store inputs at word boundary */")
oldword = 0
for j in range(howmany(bit)/8):
firstword = j * bit / 32
if(firstword > oldword):
print(" _mm256_storeu_si256(compressed + {0}, w{1});".format(oldword,oldword%2))
oldword = firstword
secondword = (j * bit + bit - 1)/32
firstshift = (j*bit) % 32
loadstr = maskfnc(" _mm256_lddqu_si256 (in + {0}) ".format(j))
if( firstword == secondword):
if(firstshift == 0):
print(" w{0} = {1};".format(firstword%2,loadstr))
else:
print(" w{0} = _mm256_or_si256(w{0},_mm256_slli_epi32({1} , {2}));".format(firstword%2,loadstr,firstshift))
else:
print(" tmp = {0};".format(loadstr))
print(" w{0} = _mm256_or_si256(w{0},_mm256_slli_epi32(tmp , {2}));".format(firstword%2,j,firstshift))
secondshift = 32-firstshift
print(" w{0} = _mm256_srli_epi32(tmp,{2});".format(secondword%2,j,secondshift))
print(" _mm256_storeu_si256(compressed + {0}, w{1});".format(secondword,secondword%2))
print("}");
print("")
print("static void avxunpackblock0(const __m256i * compressed, uint32_t * pout) {");
print(" (void) compressed;");
print(" memset(pout,0,{0});".format(howmany(0)));
print("}");
print("")
for bit in range(1,33):
print("")
print("/* we packed {0} {1}-bit values, touching {2} 256-bit words, using {3} bytes */ ".format(howmany(bit),bit,howmanywords(bit),howmanybytes(bit)))
print("static void avxunpackblock{0}(const __m256i * compressed, uint32_t * pout) {{".format(bit));
print(" /* we are going to access {0} 256-bit word{1} */ ".format(howmanywords(bit),plurial(howmanywords(bit))));
if(howmanywords(bit) == 1):
print(" __m256i w0;")
else:
print(" __m256i w0, w1;")
print(" __m256i * out = (__m256i *) pout;");
if(bit < 32): print(" const __m256i mask = _mm256_set1_epi32({0});".format((1<<bit)-1));
maskstr = " _mm256_and_si256 ( mask, {0}) "
if (bit == 32) : maskstr = " {0} " # no need
oldword = 0
print(" w0 = _mm256_lddqu_si256 (compressed);")
for j in range(howmany(bit)/8):
firstword = j * bit / 32
secondword = (j * bit + bit - 1)/32
if(secondword > oldword):
print(" w{0} = _mm256_lddqu_si256 (compressed + {1});".format(secondword%2,secondword))
oldword = secondword
firstshift = (j*bit) % 32
firstshiftstr = "_mm256_srli_epi32( w{0} , "+str(firstshift)+") "
if(firstshift == 0):
firstshiftstr =" w{0} " # no need
wfirst = firstshiftstr.format(firstword%2)
if( firstword == secondword):
if(firstshift + bit <> 32):
wfirst = maskstr.format(wfirst)
print(" _mm256_storeu_si256(out + {0}, {1});".format(j,wfirst))
else:
secondshift = (32-firstshift)
wsecond = "_mm256_slli_epi32( w{0} , {1} ) ".format((firstword+1)%2,secondshift)
wfirstorsecond = " _mm256_or_si256 ({0},{1}) ".format(wfirst,wsecond)
wfirstorsecond = maskstr.format(wfirstorsecond)
print(" _mm256_storeu_si256(out + {0},\n {1});".format(j,wfirstorsecond))
print("}");
print("")
print("static avxpackblockfnc avxfuncPackArr[] = {")
for bit in range(0,32):
print("&avxpackblock{0},".format(bit))
print("&avxpackblock32")
print("};")
print("static avxpackblockfnc avxfuncPackMaskArr[] = {")
for bit in range(0,32):
print("&avxpackblockmask{0},".format(bit))
print("&avxpackblockmask32")
print("};")
print("static avxunpackblockfnc avxfuncUnpackArr[] = {")
for bit in range(0,32):
print("&avxunpackblock{0},".format(bit))
print("&avxunpackblock32")
print("};")
print("/** code generated by avxpacking.py ends here **/")

152
cpp/simdcomp/scripts/simdfor.py Executable file
View File

@@ -0,0 +1,152 @@
#!/usr/bin/env python3
from math import ceil
print("""
/**
* Blablabla
*
*/
""");
def mask(bit):
return str((1 << bit) - 1)
for length in [32]:
print("""
static __m128i iunpackFOR0(__m128i initOffset, const __m128i * _in , uint32_t * _out) {
__m128i *out = (__m128i*)(_out);
int i;
(void) _in;
for (i = 0; i < 8; ++i) {
_mm_store_si128(out++, initOffset);
_mm_store_si128(out++, initOffset);
_mm_store_si128(out++, initOffset);
_mm_store_si128(out++, initOffset);
}
return initOffset;
}
""")
print("""
static void ipackFOR0(__m128i initOffset , const uint32_t * _in , __m128i * out ) {
(void) initOffset;
(void) _in;
(void) out;
}
""")
for bit in range(1,33):
offsetVar = " initOffset";
print("""
static void ipackFOR"""+str(bit)+"""(__m128i """+offsetVar+""", const uint32_t * _in, __m128i * out) {
const __m128i *in = (const __m128i*)(_in);
__m128i OutReg;
""");
if (bit != 32):
print(" __m128i CurrIn = _mm_load_si128(in);");
print(" __m128i InReg = _mm_sub_epi32(CurrIn, initOffset);");
else:
print(" __m128i InReg = _mm_load_si128(in);");
print(" (void) initOffset;");
inwordpointer = 0
valuecounter = 0
for k in range(ceil((length * bit) / 32)):
if(valuecounter == length): break
for x in range(inwordpointer,32,bit):
if(x!=0) :
print(" OutReg = _mm_or_si128(OutReg, _mm_slli_epi32(InReg, " + str(x) + "));");
else:
print(" OutReg = InReg; ");
if((x+bit>=32) ):
while(inwordpointer<32):
inwordpointer += bit
print(" _mm_store_si128(out, OutReg);");
print("");
if(valuecounter + 1 < length):
print(" ++out;")
inwordpointer -= 32;
if(inwordpointer>0):
print(" OutReg = _mm_srli_epi32(InReg, " + str(bit) + " - " + str(inwordpointer) + ");");
if(valuecounter + 1 < length):
print(" ++in;")
if (bit != 32):
print(" CurrIn = _mm_load_si128(in);");
print(" InReg = _mm_sub_epi32(CurrIn, initOffset);");
else:
print(" InReg = _mm_load_si128(in);");
print("");
valuecounter = valuecounter + 1
if(valuecounter == length): break
assert(valuecounter == length)
print("\n}\n\n""")
for bit in range(1,32):
offsetVar = " initOffset";
print("""\n
static __m128i iunpackFOR"""+str(bit)+"""(__m128i """+offsetVar+""", const __m128i* in, uint32_t * _out) {
""");
print(""" __m128i* out = (__m128i*)(_out);
__m128i InReg = _mm_load_si128(in);
__m128i OutReg;
__m128i tmp;
const __m128i mask = _mm_set1_epi32((1U<<"""+str(bit)+""")-1);
""");
MainText = "";
MainText += "\n";
inwordpointer = 0
valuecounter = 0
for k in range(ceil((length * bit) / 32)):
for x in range(inwordpointer,32,bit):
if(valuecounter == length): break
if (x > 0):
MainText += " tmp = _mm_srli_epi32(InReg," + str(x) +");\n";
else:
MainText += " tmp = InReg;\n";
if(x+bit<32):
MainText += " OutReg = _mm_and_si128(tmp, mask);\n";
else:
MainText += " OutReg = tmp;\n";
if((x+bit>=32) ):
while(inwordpointer<32):
inwordpointer += bit
if(valuecounter + 1 < length):
MainText += " ++in;"
MainText += " InReg = _mm_load_si128(in);\n";
inwordpointer -= 32;
if(inwordpointer>0):
MainText += " OutReg = _mm_or_si128(OutReg, _mm_and_si128(_mm_slli_epi32(InReg, " + str(bit) + "-" + str(inwordpointer) + "), mask));\n\n";
if (bit != 32):
MainText += " OutReg = _mm_add_epi32(OutReg, initOffset);\n";
MainText += " _mm_store_si128(out++, OutReg);\n\n";
MainText += "";
valuecounter = valuecounter + 1
if(valuecounter == length): break
assert(valuecounter == length)
print(MainText)
print(" return initOffset;");
print("\n}\n\n")
print("""
static __m128i iunpackFOR32(__m128i initvalue , const __m128i* in, uint32_t * _out) {
__m128i * mout = (__m128i *)_out;
__m128i invec;
size_t k;
for(k = 0; k < 128/4; ++k) {
invec = _mm_load_si128(in++);
_mm_store_si128(mout++, invec);
}
return invec;
}
""")

40
cpp/simdcomp/simdcomp.def Normal file
View File

@@ -0,0 +1,40 @@
EXPORTS
simdpack
simdpackwithoutmask
simdunpack
bits
maxbits
maxbits_length
simdmin
simdmin_length
simdmaxmin
simdmaxmin_length
simdmaxbitsd1
simdmaxbitsd1_length
simdpackd1
simdpackwithoutmaskd1
simdunpackd1
simdsearchd1
simdsearchwithlengthd1
simdselectd1
simdpackFOR
simdselectFOR
simdsearchwithlengthFOR
simdunpackFOR
simdmin_length
simdmaxmin
simdmaxmin_length
simdpack_length
simdpackFOR_length
simdunpackFOR_length
simdpack_shortlength
simdfastsetFOR
simdfastset
simdfastsetd1
simdunpack_length
simdunpack_shortlength
simdsearchwithlengthFOR
simdscand1
simdfastsetd1fromprevious
simdfastsetd1

7795
cpp/simdcomp/src/avxbitpacking.c Normal file
View File

File diff suppressed because it is too large Load Diff

14183
cpp/simdcomp/src/simdbitpacking.c Normal file
View File

File diff suppressed because it is too large Load Diff

234
cpp/simdcomp/src/simdcomputil.c Normal file
View File

@@ -0,0 +1,234 @@
/**
* This code is released under a BSD License.
*/
#include "simdcomputil.h"
#ifdef __SSE4_1__
#include <smmintrin.h>
#endif
#include <assert.h>
#define Delta(curr, prev) \
_mm_sub_epi32(curr, \
_mm_or_si128(_mm_slli_si128(curr, 4), _mm_srli_si128(prev, 12)))
/* returns the integer logarithm of v (bit width) */
uint32_t bits(const uint32_t v) {
#ifdef _MSC_VER
unsigned long answer;
if (v == 0) {
return 0;
}
_BitScanReverse(&answer, v);
return answer + 1;
#else
return v == 0 ? 0 : 32 - __builtin_clz(v); /* assume GCC-like compiler if not microsoft */
#endif
}
static uint32_t maxbitas32int(const __m128i accumulator) {
const __m128i _tmp1 = _mm_or_si128(_mm_srli_si128(accumulator, 8), accumulator); /* (A,B,C,D) xor (0,0,A,B) = (A,B,C xor A,D xor B)*/
const __m128i _tmp2 = _mm_or_si128(_mm_srli_si128(_tmp1, 4), _tmp1); /* (A,B,C xor A,D xor B) xor (0,0,0,C xor A)*/
uint32_t ans = _mm_cvtsi128_si32(_tmp2);
return bits(ans);
}
SIMDCOMP_PURE uint32_t maxbits(const uint32_t * begin) {
const __m128i* pin = (const __m128i*)(begin);
__m128i accumulator = _mm_loadu_si128(pin);
uint32_t k = 1;
for(; 4*k < SIMDBlockSize; ++k) {
__m128i newvec = _mm_loadu_si128(pin+k);
accumulator = _mm_or_si128(accumulator,newvec);
}
return maxbitas32int(accumulator);
}
static uint32_t orasint(const __m128i accumulator) {
const __m128i _tmp1 = _mm_or_si128(_mm_srli_si128(accumulator, 8), accumulator); /* (A,B,C,D) xor (0,0,A,B) = (A,B,C xor A,D xor B)*/
const __m128i _tmp2 = _mm_or_si128(_mm_srli_si128(_tmp1, 4), _tmp1); /* (A,B,C xor A,D xor B) xor (0,0,0,C xor A)*/
return _mm_cvtsi128_si32(_tmp2);
}
#ifdef __SSE4_1__
static uint32_t minasint(const __m128i accumulator) {
const __m128i _tmp1 = _mm_min_epu32(_mm_srli_si128(accumulator, 8), accumulator); /* (A,B,C,D) xor (0,0,A,B) = (A,B,C xor A,D xor B)*/
const __m128i _tmp2 = _mm_min_epu32(_mm_srli_si128(_tmp1, 4), _tmp1); /* (A,B,C xor A,D xor B) xor (0,0,0,C xor A)*/
return _mm_cvtsi128_si32(_tmp2);
}
static uint32_t maxasint(const __m128i accumulator) {
const __m128i _tmp1 = _mm_max_epu32(_mm_srli_si128(accumulator, 8), accumulator); /* (A,B,C,D) xor (0,0,A,B) = (A,B,C xor A,D xor B)*/
const __m128i _tmp2 = _mm_max_epu32(_mm_srli_si128(_tmp1, 4), _tmp1); /* (A,B,C xor A,D xor B) xor (0,0,0,C xor A)*/
return _mm_cvtsi128_si32(_tmp2);
}
uint32_t simdmin(const uint32_t * in) {
const __m128i* pin = (const __m128i*)(in);
__m128i accumulator = _mm_loadu_si128(pin);
uint32_t k = 1;
for(; 4*k < SIMDBlockSize; ++k) {
__m128i newvec = _mm_loadu_si128(pin+k);
accumulator = _mm_min_epu32(accumulator,newvec);
}
return minasint(accumulator);
}
void simdmaxmin(const uint32_t * in, uint32_t * getmin, uint32_t * getmax) {
const __m128i* pin = (const __m128i*)(in);
__m128i minaccumulator = _mm_loadu_si128(pin);
__m128i maxaccumulator = minaccumulator;
uint32_t k = 1;
for(; 4*k < SIMDBlockSize; ++k) {
__m128i newvec = _mm_loadu_si128(pin+k);
minaccumulator = _mm_min_epu32(minaccumulator,newvec);
maxaccumulator = _mm_max_epu32(maxaccumulator,newvec);
}
*getmin = minasint(minaccumulator);
*getmax = maxasint(maxaccumulator);
}
uint32_t simdmin_length(const uint32_t * in, uint32_t length) {
uint32_t currentmin = 0xFFFFFFFF;
uint32_t lengthdividedby4 = length / 4;
uint32_t offset = lengthdividedby4 * 4;
uint32_t k;
if (lengthdividedby4 > 0) {
const __m128i* pin = (const __m128i*)(in);
__m128i accumulator = _mm_loadu_si128(pin);
k = 1;
for(; 4*k < lengthdividedby4 * 4; ++k) {
__m128i newvec = _mm_loadu_si128(pin+k);
accumulator = _mm_min_epu32(accumulator,newvec);
}
currentmin = minasint(accumulator);
}
for (k = offset; k < length; ++k)
if (in[k] < currentmin)
currentmin = in[k];
return currentmin;
}
void simdmaxmin_length(const uint32_t * in, uint32_t length, uint32_t * getmin, uint32_t * getmax) {
uint32_t lengthdividedby4 = length / 4;
uint32_t offset = lengthdividedby4 * 4;
uint32_t k;
*getmin = 0xFFFFFFFF;
*getmax = 0;
if (lengthdividedby4 > 0) {
const __m128i* pin = (const __m128i*)(in);
__m128i minaccumulator = _mm_loadu_si128(pin);
__m128i maxaccumulator = minaccumulator;
k = 1;
for(; 4*k < lengthdividedby4 * 4; ++k) {
__m128i newvec = _mm_loadu_si128(pin+k);
minaccumulator = _mm_min_epu32(minaccumulator,newvec);
maxaccumulator = _mm_max_epu32(maxaccumulator,newvec);
}
*getmin = minasint(minaccumulator);
*getmax = maxasint(maxaccumulator);
}
for (k = offset; k < length; ++k) {
if (in[k] < *getmin)
*getmin = in[k];
if (in[k] > *getmax)
*getmax = in[k];
}
}
#endif
SIMDCOMP_PURE uint32_t maxbits_length(const uint32_t * in,uint32_t length) {
uint32_t k;
uint32_t lengthdividedby4 = length / 4;
uint32_t offset = lengthdividedby4 * 4;
uint32_t bigxor = 0;
if(lengthdividedby4 > 0) {
const __m128i* pin = (const __m128i*)(in);
__m128i accumulator = _mm_loadu_si128(pin);
k = 1;
for(; 4*k < 4*lengthdividedby4; ++k) {
__m128i newvec = _mm_loadu_si128(pin+k);
accumulator = _mm_or_si128(accumulator,newvec);
}
bigxor = orasint(accumulator);
}
for(k = offset; k < length; ++k)
bigxor |= in[k];
return bits(bigxor);
}
/* maxbit over 128 integers (SIMDBlockSize) with provided initial value */
uint32_t simdmaxbitsd1(uint32_t initvalue, const uint32_t * in) {
__m128i initoffset = _mm_set1_epi32 (initvalue);
const __m128i* pin = (const __m128i*)(in);
__m128i newvec = _mm_loadu_si128(pin);
__m128i accumulator = Delta(newvec , initoffset);
__m128i oldvec = newvec;
uint32_t k = 1;
for(; 4*k < SIMDBlockSize; ++k) {
newvec = _mm_loadu_si128(pin+k);
accumulator = _mm_or_si128(accumulator,Delta(newvec , oldvec));
oldvec = newvec;
}
initoffset = oldvec;
return maxbitas32int(accumulator);
}
/* maxbit over |length| integers with provided initial value */
uint32_t simdmaxbitsd1_length(uint32_t initvalue, const uint32_t * in,
uint32_t length) {
__m128i newvec;
__m128i oldvec;
__m128i initoffset;
__m128i accumulator;
const __m128i *pin;
uint32_t tmparray[4];
uint32_t k = 1;
uint32_t acc;
assert(length > 0);
pin = (const __m128i *)(in);
initoffset = _mm_set1_epi32(initvalue);
switch (length) {
case 1:
newvec = _mm_set1_epi32(in[0]);
break;
case 2:
newvec = _mm_setr_epi32(in[0], in[1], in[1], in[1]);
break;
case 3:
newvec = _mm_setr_epi32(in[0], in[1], in[2], in[2]);
break;
default:
newvec = _mm_loadu_si128(pin);
break;
}
accumulator = Delta(newvec, initoffset);
oldvec = newvec;
/* process 4 integers and build an accumulator */
while (k * 4 + 4 <= length) {
newvec = _mm_loadu_si128(pin + k);
accumulator = _mm_or_si128(accumulator, Delta(newvec, oldvec));
oldvec = newvec;
k++;
}
/* extract the accumulator as an integer */
_mm_storeu_si128((__m128i *)(tmparray), accumulator);
acc = tmparray[0] | tmparray[1] | tmparray[2] | tmparray[3];
/* now process the remaining integers */
for (k *= 4; k < length; k++)
acc |= in[k] - (k == 0 ? initvalue : in[k - 1]);
/* return the number of bits */
return bits(acc);
}

14501
cpp/simdcomp/src/simdfor.c Normal file
View File

File diff suppressed because it is too large Load Diff

24882
cpp/simdcomp/src/simdintegratedbitpacking.c Normal file
View File

File diff suppressed because it is too large Load Diff

15866
cpp/simdcomp/src/simdpackedsearch.c Normal file
View File

File diff suppressed because it is too large Load Diff

15490
cpp/simdcomp/src/simdpackedselect.c Normal file
View File

File diff suppressed because it is too large Load Diff

900
cpp/simdcomp/tests/unit.c Normal file
View File

@@ -0,0 +1,900 @@
/**
* This code is released under a BSD License.
*/
#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include "simdcomp.h"
int testshortpack() {
int bit;
size_t i;
size_t length;
__m128i * bb;
srand(0);
printf("testshortpack\n");
for (bit = 0; bit < 32; ++bit) {
const size_t N = 128;
uint32_t * data = malloc(N * sizeof(uint32_t));
uint32_t * backdata = malloc(N * sizeof(uint32_t));
uint32_t * buffer = malloc((2 * N + 1024) * sizeof(uint32_t));
for (i = 0; i < N; ++i) {
data[i] = rand() & ((1 << bit) - 1);
}
for (length = 0; length <= N; ++length) {
for (i = 0; i < N; ++i) {
backdata[i] = 0;
}
bb = simdpack_shortlength(data, length, (__m128i *) buffer,
bit);
if((bb - (__m128i *) buffer) * sizeof(__m128i) != (unsigned) simdpack_compressedbytes(length,bit)) {
printf("bug\n");
return -1;
}
simdunpack_shortlength((__m128i *) buffer, length,
backdata, bit);
for (i = 0; i < length; ++i) {
if (data[i] != backdata[i]) {
printf("bug\n");
return -1;
}
}
}
free(data);
free(backdata);
free(buffer);
}
return 0;
}
int testlongpack() {
int bit;
size_t i;
size_t length;
__m128i * bb;
srand(0);
printf("testlongpack\n");
for (bit = 0; bit < 32; ++bit) {
const size_t N = 2048;
uint32_t * data = malloc(N * sizeof(uint32_t));
uint32_t * backdata = malloc(N * sizeof(uint32_t));
uint32_t * buffer = malloc((2 * N + 1024) * sizeof(uint32_t));
for (i = 0; i < N; ++i) {
data[i] = rand() & ((1 << bit) - 1);
}
for (length = 0; length <= N; ++length) {
for (i = 0; i < N; ++i) {
backdata[i] = 0;
}
bb = simdpack_length(data, length, (__m128i *) buffer,
bit);
if((bb - (__m128i *) buffer) * sizeof(__m128i) != (unsigned) simdpack_compressedbytes(length,bit)) {
printf("bug\n");
return -1;
}
simdunpack_length((__m128i *) buffer, length,
backdata, bit);
for (i = 0; i < length; ++i) {
if (data[i] != backdata[i]) {
printf("bug\n");
return -1;
}
}
}
free(data);
free(backdata);
free(buffer);
}
return 0;
}
int testset() {
int bit;
size_t i;
const size_t N = 128;
uint32_t * data = malloc(N * sizeof(uint32_t));
uint32_t * backdata = malloc(N * sizeof(uint32_t));
uint32_t * buffer = malloc((2 * N + 1024) * sizeof(uint32_t));
srand(0);
for (bit = 0; bit < 32; ++bit) {
printf("simple set %d \n",bit);
for (i = 0; i < N; ++i) {
data[i] = rand() & ((1 << bit) - 1);
}
for (i = 0; i < N; ++i) {
backdata[i] = 0;
}
simdpack(data, (__m128i *) buffer, bit);
simdunpack((__m128i *) buffer, backdata, bit);
for (i = 0; i < N; ++i) {
if (data[i] != backdata[i]) {
printf("bug\n");
return -1;
}
}
for(i = N ; i > 0; i--) {
simdfastset((__m128i *) buffer, bit, data[N - i], i - 1);
}
simdunpack((__m128i *) buffer, backdata, bit);
for (i = 0; i < N; ++i) {
if (data[i] != backdata[N - i - 1]) {
printf("bug\n");
return -1;
}
}
simdpack(data, (__m128i *) buffer, bit);
for(i = 1 ; i <= N; i++) {
simdfastset((__m128i *) buffer, bit, data[i - 1], i - 1);
}
simdunpack((__m128i *) buffer, backdata, bit);
for (i = 0; i < N; ++i) {
if (data[i] != backdata[i]) {
printf("bug\n");
return -1;
}
}
}
free(data);
free(backdata);
free(buffer);
return 0;
}
#ifdef __SSE4_1__
int testsetd1() {
int bit;
size_t i;
uint32_t newvalue;
const size_t N = 128;
uint32_t * data = malloc(N * sizeof(uint32_t));
uint32_t * datazeroes = malloc(N * sizeof(uint32_t));
uint32_t * backdata = malloc(N * sizeof(uint32_t));
uint32_t * buffer = malloc((2 * N + 1024) * sizeof(uint32_t));
srand(0);
for (bit = 0; bit < 32; ++bit) {
printf("simple set d1 %d \n",bit);
data[0] = rand() & ((1 << bit) - 1);
datazeroes[0] = 0;
for (i = 1; i < N; ++i) {
data[i] = data[i - 1] + (rand() & ((1 << bit) - 1));
datazeroes[i] = 0;
}
for (i = 0; i < N; ++i) {
backdata[i] = 0;
}
simdpackd1(0,datazeroes, (__m128i *) buffer, bit);
for(i = 1 ; i <= N; i++) {
simdfastsetd1(0,(__m128i *) buffer, bit, data[i - 1], i - 1);
newvalue = simdselectd1(0, (const __m128i *) buffer, bit,i - 1);
if( newvalue != data[i-1] ) {
printf("bad set-select\n");
return -1;
}
}
simdunpackd1(0,(__m128i *) buffer, backdata, bit);
for (i = 0; i < N; ++i) {
if (data[i] != backdata[i])
return -1;
}
}
free(data);
free(backdata);
free(buffer);
free(datazeroes);
return 0;
}
#endif
int testsetFOR() {
int bit;
size_t i;
uint32_t newvalue;
const size_t N = 128;
uint32_t * data = malloc(N * sizeof(uint32_t));
uint32_t * datazeroes = malloc(N * sizeof(uint32_t));
uint32_t * backdata = malloc(N * sizeof(uint32_t));
uint32_t * buffer = malloc((2 * N + 1024) * sizeof(uint32_t));
srand(0);
for (bit = 0; bit < 32; ++bit) {
printf("simple set FOR %d \n",bit);
for (i = 0; i < N; ++i) {
data[i] = (rand() & ((1 << bit) - 1));
datazeroes[i] = 0;
}
for (i = 0; i < N; ++i) {
backdata[i] = 0;
}
simdpackFOR(0,datazeroes, (__m128i *) buffer, bit);
for(i = 1 ; i <= N; i++) {
simdfastsetFOR(0,(__m128i *) buffer, bit, data[i - 1], i - 1);
newvalue = simdselectFOR(0, (const __m128i *) buffer, bit,i - 1);
if( newvalue != data[i-1] ) {
printf("bad set-select\n");
return -1;
}
}
simdunpackFOR(0,(__m128i *) buffer, backdata, bit);
for (i = 0; i < N; ++i) {
if (data[i] != backdata[i])
return -1;
}
}
free(data);
free(backdata);
free(buffer);
free(datazeroes);
return 0;
}
int testshortFORpack() {
int bit;
size_t i;
__m128i * rb;
size_t length;
uint32_t offset = 7;
srand(0);
for (bit = 0; bit < 32; ++bit) {
const size_t N = 128;
uint32_t * data = malloc(N * sizeof(uint32_t));
uint32_t * backdata = malloc(N * sizeof(uint32_t));
uint32_t * buffer = malloc((2 * N + 1024) * sizeof(uint32_t));
for (i = 0; i < N; ++i) {
data[i] = (rand() & ((1 << bit) - 1)) + offset;
}
for (length = 0; length <= N; ++length) {
for (i = 0; i < N; ++i) {
backdata[i] = 0;
}
rb = simdpackFOR_length(offset,data, length, (__m128i *) buffer,
bit);
if(((rb - (__m128i *) buffer)*sizeof(__m128i)) != (unsigned) simdpackFOR_compressedbytes(length,bit)) {
return -1;
}
simdunpackFOR_length(offset,(__m128i *) buffer, length,
backdata, bit);
for (i = 0; i < length; ++i) {
if (data[i] != backdata[i])
return -1;
}
}
free(data);
free(backdata);
free(buffer);
}
return 0;
}
#ifdef __AVX2__
int testbabyavx() {
int bit;
int trial;
unsigned int i,j;
const size_t N = AVXBlockSize;
srand(0);
printf("testbabyavx\n");
printf("bit = ");
for (bit = 0; bit < 32; ++bit) {
printf(" %d ",bit);
fflush(stdout);
for(trial = 0; trial < 100; ++trial) {
uint32_t * data = malloc(N * sizeof(uint32_t)+ 64 * sizeof(uint32_t));
uint32_t * backdata = malloc(N * sizeof(uint32_t) + 64 * sizeof(uint32_t) );
__m256i * buffer = malloc((2 * N + 1024) * sizeof(uint32_t) + 32);
for (i = 0; i < N; ++i) {
data[i] = rand() & ((uint32_t)(1 << bit) - 1);
}
for (i = 0; i < N; ++i) {
backdata[i] = 0;
}
if(avxmaxbits(data) != maxbits_length(data,N)) {
printf("avxmaxbits is buggy\n");
return -1;
}
avxpackwithoutmask(data, buffer, bit);
avxunpack(buffer, backdata, bit);
for (i = 0; i < AVXBlockSize; ++i) {
if (data[i] != backdata[i]) {
printf("bug\n");
for (j = 0; j < N; ++j) {
if (data[j] != backdata[j]) {
printf("data[%d]=%d v.s. backdata[%d]=%d\n",j,data[j],j,backdata[j]);
} else {
printf("data[%d]=%d\n",j,data[j]);
}
}
return -1;
}
}
free(data);
free(backdata);
free(buffer);
}
}
printf("\n");
return 0;
}
int testavx2() {
int N = 5000 * AVXBlockSize, gap;
__m256i * buffer = malloc(AVXBlockSize * sizeof(uint32_t));
uint32_t * datain = malloc(N * sizeof(uint32_t));
uint32_t * backbuffer = malloc(AVXBlockSize * sizeof(uint32_t));
for (gap = 1; gap <= 387420489; gap *= 3) {
int k;
printf(" gap = %u \n", gap);
for (k = 0; k < N; ++k)
datain[k] = k * gap;
for (k = 0; k * AVXBlockSize < N; ++k) {
/*
First part works for general arrays (sorted or unsorted)
*/
int j;
/* we compute the bit width */
const uint32_t b = avxmaxbits(datain + k * AVXBlockSize);
if(avxmaxbits(datain + k * AVXBlockSize) != maxbits_length(datain + k * AVXBlockSize,AVXBlockSize)) {
printf("avxmaxbits is buggy %d %d \n",
avxmaxbits(datain + k * AVXBlockSize),
maxbits_length(datain + k * AVXBlockSize,AVXBlockSize));
return -1;
}
printf("bit width = %d\n",b);
/* we read 256 integers at "datain + k * AVXBlockSize" and
write b 256-bit vectors at "buffer" */
avxpackwithoutmask(datain + k * AVXBlockSize, buffer, b);
/* we read back b1 128-bit vectors at "buffer" and write 128 integers at backbuffer */
avxunpack(buffer, backbuffer, b);/* uncompressed */
for (j = 0; j < AVXBlockSize; ++j) {
if (backbuffer[j] != datain[k * AVXBlockSize + j]) {
int i;
printf("bug in avxpack\n");
for(i = 0; i < AVXBlockSize; ++i) {
printf("data[%d]=%d got back %d %s\n",i,
datain[k * AVXBlockSize + i],backbuffer[i],
datain[k * AVXBlockSize + i]!=backbuffer[i]?"bug":"");
}
return -2;
}
}
}
}
free(buffer);
free(datain);
free(backbuffer);
printf("Code looks good.\n");
return 0;
}
#endif /* avx2 */
int test() {
int N = 5000 * SIMDBlockSize, gap;
__m128i * buffer = malloc(SIMDBlockSize * sizeof(uint32_t));
uint32_t * datain = malloc(N * sizeof(uint32_t));
uint32_t * backbuffer = malloc(SIMDBlockSize * sizeof(uint32_t));
for (gap = 1; gap <= 387420489; gap *= 3) {
int k;
printf(" gap = %u \n", gap);
for (k = 0; k < N; ++k)
datain[k] = k * gap;
for (k = 0; k * SIMDBlockSize < N; ++k) {
/*
First part works for general arrays (sorted or unsorted)
*/
int j;
/* we compute the bit width */
const uint32_t b = maxbits(datain + k * SIMDBlockSize);
/* we read 128 integers at "datain + k * SIMDBlockSize" and
write b 128-bit vectors at "buffer" */
simdpackwithoutmask(datain + k * SIMDBlockSize, buffer, b);
/* we read back b1 128-bit vectors at "buffer" and write 128 integers at backbuffer */
simdunpack(buffer, backbuffer, b);/* uncompressed */
for (j = 0; j < SIMDBlockSize; ++j) {
if (backbuffer[j] != datain[k * SIMDBlockSize + j]) {
printf("bug in simdpack\n");
return -2;
}
}
{
/*
next part assumes that the data is sorted (uses differential coding)
*/
uint32_t offset = 0;
/* we compute the bit width */
const uint32_t b1 = simdmaxbitsd1(offset,
datain + k * SIMDBlockSize);
/* we read 128 integers at "datain + k * SIMDBlockSize" and
write b1 128-bit vectors at "buffer" */
simdpackwithoutmaskd1(offset, datain + k * SIMDBlockSize, buffer,
b1);
/* we read back b1 128-bit vectors at "buffer" and write 128 integers at backbuffer */
simdunpackd1(offset, buffer, backbuffer, b1);
for (j = 0; j < SIMDBlockSize; ++j) {
if (backbuffer[j] != datain[k * SIMDBlockSize + j]) {
printf("bug in simdpack d1\n");
return -3;
}
}
offset = datain[k * SIMDBlockSize + SIMDBlockSize - 1];
}
}
}
free(buffer);
free(datain);
free(backbuffer);
printf("Code looks good.\n");
return 0;
}
#ifdef __SSE4_1__
int testFOR() {
int N = 5000 * SIMDBlockSize, gap;
__m128i * buffer = malloc(SIMDBlockSize * sizeof(uint32_t));
uint32_t * datain = malloc(N * sizeof(uint32_t));
uint32_t * backbuffer = malloc(SIMDBlockSize * sizeof(uint32_t));
uint32_t tmax, tmin, tb;
for (gap = 1; gap <= 387420489; gap *= 2) {
int k;
printf(" gap = %u \n", gap);
for (k = 0; k < N; ++k)
datain[k] = k * gap;
for (k = 0; k * SIMDBlockSize < N; ++k) {
int j;
simdmaxmin_length(datain + k * SIMDBlockSize,SIMDBlockSize,&tmin,&tmax);
/* we compute the bit width */
tb = bits(tmax - tmin);
/* we read 128 integers at "datain + k * SIMDBlockSize" and
write b 128-bit vectors at "buffer" */
simdpackFOR(tmin,datain + k * SIMDBlockSize, buffer, tb);
for (j = 0; j < SIMDBlockSize; ++j) {
uint32_t selectedvalue = simdselectFOR(tmin,buffer,tb,j);
if (selectedvalue != datain[k * SIMDBlockSize + j]) {
printf("bug in simdselectFOR\n");
return -3;
}
}
/* we read back b1 128-bit vectors at "buffer" and write 128 integers at backbuffer */
simdunpackFOR(tmin,buffer, backbuffer, tb);/* uncompressed */
for (j = 0; j < SIMDBlockSize; ++j) {
if (backbuffer[j] != datain[k * SIMDBlockSize + j]) {
printf("bug in simdpackFOR\n");
return -2;
}
}
}
}
free(buffer);
free(datain);
free(backbuffer);
printf("Code looks good.\n");
return 0;
}
#endif
#define MAX 300
int test_simdmaxbitsd1_length() {
uint32_t result, buffer[MAX + 1];
int i, j;
memset(&buffer[0], 0xff, sizeof(buffer));
/* this test creates buffers of different length; each buffer is
* initialized to result in the following deltas:
* length 1: 2
* length 2: 1 2
* length 3: 1 1 2
* length 4: 1 1 1 2
* length 5: 1 1 1 1 2
* etc. Each sequence's "maxbits" is 2. */
for (i = 0; i < MAX; i++) {
for (j = 0; j < i; j++)
buffer[j] = j + 1;
buffer[i] = i + 2;
result = simdmaxbitsd1_length(0, &buffer[0], i + 1);
if (result != 2) {
printf("simdmaxbitsd1_length: unexpected result %u in loop %d\n",
result, i);
return -1;
}
}
printf("simdmaxbitsd1_length: ok\n");
return 0;
}
int uint32_cmp(const void *a, const void *b)
{
const uint32_t *ia = (const uint32_t *)a;
const uint32_t *ib = (const uint32_t *)b;
if(*ia < *ib)
return -1;
else if (*ia > *ib)
return 1;
return 0;
}
#ifdef __SSE4_1__
int test_simdpackedsearch() {
uint32_t buffer[128];
uint32_t result = 0;
int b, i;
uint32_t init = 0;
__m128i initial = _mm_set1_epi32(init);
/* initialize the buffer */
for (i = 0; i < 128; i++)
buffer[i] = (uint32_t)(i + 1);
/* this test creates delta encoded buffers with different bits, then
* performs lower bound searches for each key */
for (b = 1; b <= 32; b++) {
uint32_t out[128];
/* delta-encode to 'i' bits */
simdpackwithoutmaskd1(init, buffer, (__m128i *)out, b);
initial = _mm_setzero_si128();
printf("simdsearchd1: %d bits\n", b);
/* now perform the searches */
initial = _mm_set1_epi32(init);
assert(simdsearchd1(&initial, (__m128i *)out, b, 0, &result) == 0);
assert(result > 0);
for (i = 1; i <= 128; i++) {
initial = _mm_set1_epi32(init);
assert(simdsearchd1(&initial, (__m128i *)out, b,
(uint32_t)i, &result) == i - 1);
assert(result == (unsigned)i);
}
initial = _mm_set1_epi32(init);
assert(simdsearchd1(&initial, (__m128i *)out, b, 200, &result)
== 128);
assert(result > 200);
}
printf("simdsearchd1: ok\n");
return 0;
}
int test_simdpackedsearchFOR() {
uint32_t buffer[128];
uint32_t result = 0;
int b;
uint32_t i;
uint32_t maxv, tmin, tmax, tb;
uint32_t out[128];
/* this test creates delta encoded buffers with different bits, then
* performs lower bound searches for each key */
for (b = 1; b <= 32; b++) {
/* initialize the buffer */
maxv = (b == 32)
? 0xFFFFFFFF
: ((1U<<b) - 1);
for (i = 0; i < 128; i++)
buffer[i] = maxv * (i + 1) / 128;
simdmaxmin_length(buffer,SIMDBlockSize,&tmin,&tmax);
/* we compute the bit width */
tb = bits(tmax - tmin);
/* delta-encode to 'i' bits */
simdpackFOR(tmin, buffer, (__m128i *)out, tb);
printf("simdsearchd1: %d bits\n", b);
/* now perform the searches */
for (i = 0; i < 128; i++) {
assert(buffer[i] == simdselectFOR(tmin, (__m128i *)out, tb,i));
}
for (i = 0; i < 128; i++) {
int x = simdsearchwithlengthFOR(tmin, (__m128i *)out, tb,
128,buffer[i], &result) ;
assert(simdselectFOR(tmin, (__m128i *)out, tb,x) == buffer[x]);
assert(simdselectFOR(tmin, (__m128i *)out, tb,x) == result);
assert(buffer[x] == result);
assert(result == buffer[i]);
assert(buffer[x] == buffer[i]);
}
}
printf("simdsearchFOR: ok\n");
return 0;
}
int test_simdpackedsearch_advanced() {
uint32_t buffer[128];
uint32_t backbuffer[128];
uint32_t out[128];
uint32_t result = 0;
uint32_t b, i;
uint32_t init = 0;
__m128i initial = _mm_set1_epi32(init);
/* this test creates delta encoded buffers with different bits, then
* performs lower bound searches for each key */
for (b = 0; b <= 32; b++) {
uint32_t prev = init;
/* initialize the buffer */
for (i = 0; i < 128; i++) {
buffer[i] = ((uint32_t)(1431655765 * i + 0xFFFFFFFF)) ;
if(b < 32) buffer[i] %= (1<<b);
}
qsort(buffer,128, sizeof(uint32_t), uint32_cmp);
for (i = 0; i < 128; i++) {
buffer[i] = buffer[i] + prev;
prev = buffer[i];
}
for (i = 1; i < 128; i++) {
if(buffer[i] < buffer[i-1] )
buffer[i] = buffer[i-1];
}
assert(simdmaxbitsd1(init, buffer)<=b);
for (i = 0; i < 128; i++) {
out[i] = 0; /* memset would do too */
}
/* delta-encode to 'i' bits */
simdpackwithoutmaskd1(init, buffer, (__m128i *)out, b);
simdunpackd1(init, (__m128i *)out, backbuffer, b);
for (i = 0; i < 128; i++) {
assert(buffer[i] == backbuffer[i]);
}
printf("advanced simdsearchd1: %d bits\n", b);
for (i = 0; i < 128; i++) {
int pos;
initial = _mm_set1_epi32(init);
pos = simdsearchd1(&initial, (__m128i *)out, b,
buffer[i], &result);
assert(pos == simdsearchwithlengthd1(init, (__m128i *)out, b, 128,
buffer[i], &result));
assert(buffer[pos] == buffer[i]);
if(pos > 0)
assert(buffer[pos - 1] < buffer[i]);
assert(result == buffer[i]);
}
for (i = 0; i < 128; i++) {
int pos;
if(buffer[i] == 0) continue;
initial = _mm_set1_epi32(init);
pos = simdsearchd1(&initial, (__m128i *)out, b,
buffer[i] - 1, &result);
assert(pos == simdsearchwithlengthd1(init, (__m128i *)out, b, 128,
buffer[i] - 1, &result));
assert(buffer[pos] >= buffer[i] - 1);
if(pos > 0)
assert(buffer[pos - 1] < buffer[i] - 1);
assert(result == buffer[pos]);
}
for (i = 0; i < 128; i++) {
int pos;
if (buffer[i] + 1 == 0)
continue;
initial = _mm_set1_epi32(init);
pos = simdsearchd1(&initial, (__m128i *) out, b,
buffer[i] + 1, &result);
assert(pos == simdsearchwithlengthd1(init, (__m128i *)out, b, 128,
buffer[i] + 1, &result));
if(pos == 128) {
assert(buffer[i] == buffer[127]);
} else {
assert(buffer[pos] >= buffer[i] + 1);
if (pos > 0)
assert(buffer[pos - 1] < buffer[i] + 1);
assert(result == buffer[pos]);
}
}
}
printf("advanced simdsearchd1: ok\n");
return 0;
}
int test_simdpackedselect() {
uint32_t buffer[128];
uint32_t initial = 33;
int b, i;
/* initialize the buffer */
for (i = 0; i < 128; i++)
buffer[i] = (uint32_t)(initial + i);
/* this test creates delta encoded buffers with different bits, then
* performs lower bound searches for each key */
for (b = 1; b <= 32; b++) {
uint32_t out[128];
/* delta-encode to 'i' bits */
simdpackwithoutmaskd1(initial, buffer, (__m128i *)out, b);
printf("simdselectd1: %d bits\n", b);
/* now perform the searches */
for (i = 0; i < 128; i++) {
assert(simdselectd1(initial, (__m128i *)out, b, (uint32_t)i)
== initial + i);
}
}
printf("simdselectd1: ok\n");
return 0;
}
int test_simdpackedselect_advanced() {
uint32_t buffer[128];
uint32_t initial = 33;
uint32_t b;
int i;
/* this test creates delta encoded buffers with different bits, then
* performs lower bound searches for each key */
for (b = 0; b <= 32; b++) {
uint32_t prev = initial;
uint32_t out[128];
/* initialize the buffer */
for (i = 0; i < 128; i++) {
buffer[i] = ((uint32_t)(165576 * i)) ;
if(b < 32) buffer[i] %= (1<<b);
}
for (i = 0; i < 128; i++) {
buffer[i] = buffer[i] + prev;
prev = buffer[i];
}
for (i = 1; i < 128; i++) {
if(buffer[i] < buffer[i-1] )
buffer[i] = buffer[i-1];
}
assert(simdmaxbitsd1(initial, buffer)<=b);
for (i = 0; i < 128; i++) {
out[i] = 0; /* memset would do too */
}
/* delta-encode to 'i' bits */
simdpackwithoutmaskd1(initial, buffer, (__m128i *)out, b);
printf("simdselectd1: %d bits\n", b);
/* now perform the searches */
for (i = 0; i < 128; i++) {
uint32_t valretrieved = simdselectd1(initial, (__m128i *)out, b, (uint32_t)i);
assert(valretrieved == buffer[i]);
}
}
printf("advanced simdselectd1: ok\n");
return 0;
}
#endif
int main() {
int r;
r = testsetFOR();
if (r) {
printf("test failure 1\n");
return r;
}
#ifdef __SSE4_1__
r = testsetd1();
if (r) {
printf("test failure 2\n");
return r;
}
#endif
r = testset();
if (r) {
printf("test failure 3\n");
return r;
}
r = testshortFORpack();
if (r) {
printf("test failure 4\n");
return r;
}
r = testshortpack();
if (r) {
printf("test failure 5\n");
return r;
}
r = testlongpack();
if (r) {
printf("test failure 6\n");
return r;
}
#ifdef __SSE4_1__
r = test_simdpackedsearchFOR();
if (r) {
printf("test failure 7\n");
return r;
}
r = testFOR();
if (r) {
printf("test failure 8\n");
return r;
}
#endif
#ifdef __AVX2__
r= testbabyavx();
if (r) {
printf("test failure baby avx\n");
return r;
}
r = testavx2();
if (r) {
printf("test failure 9 avx\n");
return r;
}
#endif
r = test();
if (r) {
printf("test failure 9\n");
return r;
}
r = test_simdmaxbitsd1_length();
if (r) {
printf("test failure 10\n");
return r;
}
#ifdef __SSE4_1__
r = test_simdpackedsearch();
if (r) {
printf("test failure 11\n");
return r;
}
r = test_simdpackedsearch_advanced();
if (r) {
printf("test failure 12\n");
return r;
}
r = test_simdpackedselect();
if (r) {
printf("test failure 13\n");
return r;
}
r = test_simdpackedselect_advanced();
if (r) {
printf("test failure 14\n");
return r;
}
#endif
printf("All tests OK!\n");
return 0;
}

102
cpp/simdcomp/tests/unit_chars.c Normal file
View File

@@ -0,0 +1,102 @@
/**
* This code is released under a BSD License.
*/
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include "simdcomp.h"
#define get_random_char() (uint8_t)(rand() % 256);
int main() {
int N = 5000 * SIMDBlockSize, gap;
__m128i * buffer = malloc(SIMDBlockSize * sizeof(uint32_t));
uint32_t * datain = malloc(N * sizeof(uint32_t));
uint32_t * backbuffer = malloc(SIMDBlockSize * sizeof(uint32_t));
srand(time(NULL));
for (gap = 1; gap <= 387420489; gap *= 3) {
int k;
printf(" gap = %u \n", gap);
/* simulate some random character string, don't care about endiannes */
for (k = 0; k < N; ++k) {
uint8_t _tmp[4];
_tmp[0] = get_random_char();
_tmp[1] = get_random_char();
_tmp[2] = get_random_char();
_tmp[3] = get_random_char();
memmove(&datain[k], _tmp, 4);
}
for (k = 0; k * SIMDBlockSize < N; ++k) {
/*
First part works for general arrays (sorted or unsorted)
*/
int j;
/* we compute the bit width */
const uint32_t b = maxbits(datain + k * SIMDBlockSize);
/* we read 128 integers at "datain + k * SIMDBlockSize" and
write b 128-bit vectors at "buffer" */
simdpackwithoutmask(datain + k * SIMDBlockSize, buffer, b);
/* we read back b1 128-bit vectors at "buffer" and write 128 integers at backbuffer */
simdunpack(buffer, backbuffer, b);/* uncompressed */
for (j = 0; j < SIMDBlockSize; ++j) {
uint8_t chars_back[4];
uint8_t chars_in[4];
memmove(chars_back, &backbuffer[j], 4);
memmove(chars_in, &datain[k * SIMDBlockSize + j], 4);
if (chars_in[0] != chars_back[0]
|| chars_in[1] != chars_back[1]
|| chars_in[2] != chars_back[2]
|| chars_in[3] != chars_back[3]) {
printf("bug in simdpack\n");
return -2;
}
}
{
/*
next part assumes that the data is sorted (uses differential coding)
*/
uint32_t offset = 0;
/* we compute the bit width */
const uint32_t b1 = simdmaxbitsd1(offset,
datain + k * SIMDBlockSize);
/* we read 128 integers at "datain + k * SIMDBlockSize" and
write b1 128-bit vectors at "buffer" */
simdpackwithoutmaskd1(offset, datain + k * SIMDBlockSize, buffer,
b1);
/* we read back b1 128-bit vectors at "buffer" and write 128 integers at backbuffer */
simdunpackd1(offset, buffer, backbuffer, b1);
for (j = 0; j < SIMDBlockSize; ++j) {
uint8_t chars_back[4];
uint8_t chars_in[4];
memmove(chars_back, &backbuffer[j], 4);
memmove(chars_in, &datain[k * SIMDBlockSize + j], 4);
if (chars_in[0] != chars_back[0]
|| chars_in[1] != chars_back[1]
|| chars_in[2] != chars_back[2]
|| chars_in[3] != chars_back[3]) {
printf("bug in simdpack\n");
return -3;
}
}
offset = datain[k * SIMDBlockSize + SIMDBlockSize - 1];
}
}
}
free(buffer);
free(datain);
free(backbuffer);
printf("Code looks good.\n");
return 0;
}

42
cpp/simdcomp_wrapper.c vendored Normal file
View File

@@ -0,0 +1,42 @@
#include "simdcomp.h"
#include "simdcomputil.h"
// assumes datain has a size of 128 uint32
// and that buffer is large enough to host the data.
size_t compress_sorted(
const uint32_t* datain,
uint8_t* output,
const uint32_t offset) {
const uint32_t b = simdmaxbitsd1(offset, datain);
*output++ = b;
simdpackwithoutmaskd1(offset, datain, (__m128i *) output, b);
return 1 + b * sizeof(__m128i);
}
// assumes datain has a size of 128 uint32
// and that buffer is large enough to host the data.
size_t uncompress_sorted(
const uint8_t* compressed_data,
uint32_t* output,
uint32_t offset) {
const uint32_t b = *compressed_data++;
simdunpackd1(offset, (__m128i *)compressed_data, output, b);
return 1 + b * sizeof(__m128i);
}
size_t compress_unsorted(
const uint32_t* datain,
uint8_t* output) {
const uint32_t b = maxbits(datain);
*output++ = b;
simdpackwithoutmask(datain, (__m128i *) output, b);
return 1 + b * sizeof(__m128i);
}
size_t uncompress_unsorted(
const uint8_t* compressed_data,
uint32_t* output) {
const uint32_t b = *compressed_data++;
simdunpack((__m128i *)compressed_data, output, b);
return 1 + b * sizeof(__m128i);
}

48
cpp/simdcomp_wrapper.cpp vendored
View File

@@ -1,48 +0,0 @@
#include <stdio.h>
#include <time.h>
#include <stdlib.h>
#include "simdcomp.h"
#include "simdcomputil.h"
extern "C" {
// assumes datain has a size of 128 uint32
// and that buffer is large enough to host the data.
size_t compress_sorted_cpp(
const uint32_t* datain,
uint8_t* output,
const uint32_t offset) {
const uint32_t b = simdmaxbitsd1(offset, datain);
*output++ = b;
simdpackwithoutmaskd1(offset, datain, (__m128i *) output, b);
return 1 + b * sizeof(__m128i);;
}
// assumes datain has a size of 128 uint32
// and that buffer is large enough to host the data.
size_t uncompress_sorted_cpp(
const uint8_t* compressed_data,
uint32_t* output,
uint32_t offset) {
const uint32_t b = *compressed_data++;
simdunpackd1(offset, (__m128i *)compressed_data, output, b);
return 1 + b * sizeof(__m128i);
}
size_t compress_unsorted_cpp(
const uint32_t* datain,
uint8_t* output) {
const uint32_t b = maxbits(datain);
*output++ = b;
simdpackwithoutmask(datain, (__m128i *) output, b);
return 1 + b * sizeof(__m128i);;
}
size_t uncompress_unsorted_cpp(
const uint8_t* compressed_data,
uint32_t* output) {
const uint32_t b = *compressed_data++;
simdunpack((__m128i *)compressed_data, output, b);
return 1 + b * sizeof(__m128i);
}
}

32
cpp/streamvbyte/.gitignore vendored Normal file
View File

@@ -0,0 +1,32 @@
# Object files
*.o
*.ko
*.obj
*.elf
# Precompiled Headers
*.gch
*.pch
# Libraries
*.lib
*.a
*.la
*.lo
# Shared objects (inc. Windows DLLs)
*.dll
*.so
*.so.*
*.dylib
# Executables
*.exe
*.out
*.app
*.i*86
*.x86_64
*.hex
# Debug files
*.dSYM/

7
cpp/streamvbyte/.travis.yml Normal file
View File

@@ -0,0 +1,7 @@
language: c
sudo: false
compiler:
- gcc
- clang
script: make && ./unit

202
cpp/streamvbyte/LICENSE Normal file
View File

@@ -0,0 +1,202 @@
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "{}"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright {yyyy} {name of copyright owner}
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

60
cpp/streamvbyte/README.md Normal file
View File

@@ -0,0 +1,60 @@
streamvbyte
===========
[![Build Status](https://travis-ci.org/lemire/streamvbyte.png)](https://travis-ci.org/lemire/streamvbyte)
StreamVByte is a new integer compression technique that applies SIMD instructions (vectorization) to
Google's Group Varint approach. The net result is faster than other byte-oriented compression
techniques.
The approach is patent-free, the code is available under the Apache License.
It includes fast differential coding.
It assumes a recent Intel processor (e.g., haswell or better) .
The code should build using most standard-compliant C99 compilers. The provided makefile
expects a Linux-like system.
Usage:
make
./unit
See example.c for an example.
Short code sample:
```C
// suppose that datain is an array of uint32_t integers
size_t compsize = streamvbyte_encode(datain, N, compressedbuffer); // encoding
// here the result is stored in compressedbuffer using compsize bytes
streamvbyte_decode(compressedbuffer, recovdata, N); // decoding (fast)
```
If the values are sorted, then it might be preferable to use differential coding:
```C
// suppose that datain is an array of uint32_t integers
size_t compsize = streamvbyte_delta_encode(datain, N, compressedbuffer,0); // encoding
// here the result is stored in compressedbuffer using compsize bytes
streamvbyte_delta_decode(compressedbuffer, recovdata, N,0); // decoding (fast)
```
You have to know how many integers were coded when you decompress. You can store this
information along with the compressed stream.
See also
--------
* SIMDCompressionAndIntersection: A C++ library to compress and intersect sorted lists of integers using SIMD instructions https://github.com/lemire/SIMDCompressionAndIntersect
* The FastPFOR C++ library : Fast integer compression https://github.com/lemire/FastPFor
* High-performance dictionary coding https://github.com/lemire/dictionary
* LittleIntPacker: C library to pack and unpack short arrays of integers as fast as possible https://github.com/lemire/LittleIntPacker
* The SIMDComp library: A simple C library for compressing lists of integers using binary packing https://github.com/lemire/simdcomp
* MaskedVByte: Fast decoder for VByte-compressed integers https://github.com/lemire/MaskedVByte
* CSharpFastPFOR: A C# integer compression library https://github.com/Genbox/CSharpFastPFOR
* JavaFastPFOR: A java integer compression library https://github.com/lemire/JavaFastPFOR
* Encoding: Integer Compression Libraries for Go https://github.com/zhenjl/encoding
* FrameOfReference is a C++ library dedicated to frame-of-reference (FOR) compression: https://github.com/lemire/FrameOfReference
* libvbyte: A fast implementation for varbyte 32bit/64bit integer compression https://github.com/cruppstahl/libvbyte
* TurboPFor is a C library that offers lots of interesting optimizations. Well worth checking! (GPL license) https://github.com/powturbo/TurboPFor
* Oroch is a C++ library that offers a usable API (MIT license) https://github.com/ademakov/Oroch

24
cpp/streamvbyte/example.c Normal file
View File

@@ -0,0 +1,24 @@
#include <stdio.h>
#include <stdlib.h>
#include <assert.h>
#include "streamvbyte.h"
int main() {
int N = 5000;
uint32_t * datain = malloc(N * sizeof(uint32_t));
uint8_t * compressedbuffer = malloc(N * sizeof(uint32_t));
uint32_t * recovdata = malloc(N * sizeof(uint32_t));
for (int k = 0; k < N; ++k)
datain[k] = 120;
size_t compsize = streamvbyte_encode(datain, N, compressedbuffer); // encoding
// here the result is stored in compressedbuffer using compsize bytes
size_t compsize2 = streamvbyte_decode(compressedbuffer, recovdata,
N); // decoding (fast)
assert(compsize == compsize2);
free(datain);
free(compressedbuffer);
free(recovdata);
printf("Compressed %d integers down to %d bytes.\n",N,(int) compsize);
return 0;
}

19
cpp/streamvbyte/include/streamvbyte.h Normal file
View File

@@ -0,0 +1,19 @@
#ifndef VARINTDECODE_H_
#define VARINTDECODE_H_
#define __STDC_FORMAT_MACROS
#include <inttypes.h>
#include <stdint.h>// please use a C99-compatible compiler
#include <stddef.h>
// Encode an array of a given length read from in to bout in varint format.
// Returns the number of bytes written.
size_t streamvbyte_encode(const uint32_t *in, uint32_t length, uint8_t *out);
// Read "length" 32-bit integers in varint format from in, storing the result in out.
// Returns the number of bytes read.
size_t streamvbyte_decode(const uint8_t* in, uint32_t* out, uint32_t length);
#endif /* VARINTDECODE_H_ */

24
cpp/streamvbyte/include/streamvbytedelta.h Normal file
View File

@@ -0,0 +1,24 @@
/*
* streamvbytedelta.h
*
* Created on: Apr 14, 2016
* Author: lemire
*/
#ifndef INCLUDE_STREAMVBYTEDELTA_H_
#define INCLUDE_STREAMVBYTEDELTA_H_
// Encode an array of a given length read from in to bout in StreamVByte format.
// Returns the number of bytes written.
// this version uses differential coding (coding differences between values) starting at prev (you can often set prev to zero)
size_t streamvbyte_delta_encode(const uint32_t *in, uint32_t length, uint8_t *out, uint32_t prev);
// Read "length" 32-bit integers in StreamVByte format from in, storing the result in out.
// Returns the number of bytes read.
// this version uses differential coding (coding differences between values) starting at prev (you can often set prev to zero)
size_t streamvbyte_delta_decode(const uint8_t* in, uint32_t* out, uint32_t length, uint32_t prev);
#endif /* INCLUDE_STREAMVBYTEDELTA_H_ */

58
cpp/streamvbyte/makefile Normal file
View File

@@ -0,0 +1,58 @@
# minimalist makefile
.SUFFIXES:
#
.SUFFIXES: .cpp .o .c .h
CFLAGS = -fPIC -march=native -std=c99 -O3 -Wall -Wextra -pedantic -Wshadow
LDFLAGS = -shared
LIBNAME=libstreamvbyte.so.0.0.1
all: unit $(LIBNAME)
test:
./unit
install: $(OBJECTS)
cp $(LIBNAME) /usr/local/lib
ln -s /usr/local/lib/$(LIBNAME) /usr/local/lib/libstreamvbyte.so
ldconfig
cp $(HEADERS) /usr/local/include
HEADERS=./include/streamvbyte.h ./include/streamvbytedelta.h
uninstall:
for h in $(HEADERS) ; do rm /usr/local/$$h; done
rm /usr/local/lib/$(LIBNAME)
rm /usr/local/lib/libstreamvbyte.so
ldconfig
OBJECTS= streamvbyte.o streamvbytedelta.o
streamvbytedelta.o: ./src/streamvbytedelta.c $(HEADERS)
$(CC) $(CFLAGS) -c ./src/streamvbytedelta.c -Iinclude
streamvbyte.o: ./src/streamvbyte.c $(HEADERS)
$(CC) $(CFLAGS) -c ./src/streamvbyte.c -Iinclude
$(LIBNAME): $(OBJECTS)
$(CC) $(CFLAGS) -o $(LIBNAME) $(OBJECTS) $(LDFLAGS)
example: ./example.c $(HEADERS) $(OBJECTS)
$(CC) $(CFLAGS) -o example ./example.c -Iinclude $(OBJECTS)
unit: ./tests/unit.c $(HEADERS) $(OBJECTS)
$(CC) $(CFLAGS) -o unit ./tests/unit.c -Iinclude $(OBJECTS)
dynunit: ./tests/unit.c $(HEADERS) $(LIBNAME)
$(CC) $(CFLAGS) -o dynunit ./tests/unit.c -Iinclude -lstreamvbyte
clean:
rm -f unit *.o $(LIBNAME) example

495
cpp/streamvbyte/src/streamvbyte.c Normal file
View File

@@ -0,0 +1,495 @@
#include "streamvbyte.h"
#if defined(_MSC_VER)
/* Microsoft C/C++-compatible compiler */
#include <intrin.h>
#elif defined(__GNUC__) && (defined(__x86_64__) || defined(__i386__))
/* GCC-compatible compiler, targeting x86/x86-64 */
#include <x86intrin.h>
#elif defined(__GNUC__) && defined(__ARM_NEON__)
/* GCC-compatible compiler, targeting ARM with NEON */
#include <arm_neon.h>
#elif defined(__GNUC__) && defined(__IWMMXT__)
/* GCC-compatible compiler, targeting ARM with WMMX */
#include <mmintrin.h>
#elif (defined(__GNUC__) || defined(__xlC__)) && (defined(__VEC__) || defined(__ALTIVEC__))
/* XLC or GCC-compatible compiler, targeting PowerPC with VMX/VSX */
#include <altivec.h>
#elif defined(__GNUC__) && defined(__SPE__)
/* GCC-compatible compiler, targeting PowerPC with SPE */
#include <spe.h>
#endif
static uint8_t lengthTable[256] = { 4, 5, 6, 7, 5, 6, 7, 8, 6, 7, 8, 9, 7, 8, 9,
10, 5, 6, 7, 8, 6, 7, 8, 9, 7, 8, 9, 10, 8, 9, 10, 11, 6, 7, 8, 9, 7, 8,
9, 10, 8, 9, 10, 11, 9, 10, 11, 12, 7, 8, 9, 10, 8, 9, 10, 11, 9, 10,
11, 12, 10, 11, 12, 13, 5, 6, 7, 8, 6, 7, 8, 9, 7, 8, 9, 10, 8, 9, 10,
11, 6, 7, 8, 9, 7, 8, 9, 10, 8, 9, 10, 11, 9, 10, 11, 12, 7, 8, 9, 10,
8, 9, 10, 11, 9, 10, 11, 12, 10, 11, 12, 13, 8, 9, 10, 11, 9, 10, 11,
12, 10, 11, 12, 13, 11, 12, 13, 14, 6, 7, 8, 9, 7, 8, 9, 10, 8, 9, 10,
11, 9, 10, 11, 12, 7, 8, 9, 10, 8, 9, 10, 11, 9, 10, 11, 12, 10, 11, 12,
13, 8, 9, 10, 11, 9, 10, 11, 12, 10, 11, 12, 13, 11, 12, 13, 14, 9, 10,
11, 12, 10, 11, 12, 13, 11, 12, 13, 14, 12, 13, 14, 15, 7, 8, 9, 10, 8,
9, 10, 11, 9, 10, 11, 12, 10, 11, 12, 13, 8, 9, 10, 11, 9, 10, 11, 12,
10, 11, 12, 13, 11, 12, 13, 14, 9, 10, 11, 12, 10, 11, 12, 13, 11, 12,
13, 14, 12, 13, 14, 15, 10, 11, 12, 13, 11, 12, 13, 14, 12, 13, 14, 15,
13, 14, 15, 16 };
static uint8_t shuffleTable[256][16] = { { 0, -1, -1, -1, 1, -1, -1, -1, 2, -1,
-1, -1, 3, -1, -1, -1 }, // 1111
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, -1, -1, -1, 4, -1, -1, -1 }, // 2111
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, -1, -1, -1, 5, -1, -1, -1 }, // 3111
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, -1, -1, -1, 6, -1, -1, -1 }, // 4111
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, -1, -1, -1, 4, -1, -1, -1 }, // 1211
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, -1, -1, -1, 5, -1, -1, -1 }, // 2211
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, -1, -1, -1, 6, -1, -1, -1 }, // 3211
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, -1, -1, -1, 7, -1, -1, -1 }, // 4211
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, -1, -1, -1, 5, -1, -1, -1 }, // 1311
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, -1, -1, -1, 6, -1, -1, -1 }, // 2311
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, -1, -1, -1, 7, -1, -1, -1 }, // 3311
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, -1, -1, -1, 8, -1, -1, -1 }, // 4311
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, -1, -1, -1, 6, -1, -1, -1 }, // 1411
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, -1, -1, -1, 7, -1, -1, -1 }, // 2411
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, -1, -1, -1, 8, -1, -1, -1 }, // 3411
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, -1, -1, -1, 9, -1, -1, -1 }, // 4411
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, -1, -1, 4, -1, -1, -1 }, // 1121
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, -1, -1, 5, -1, -1, -1 }, // 2121
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, -1, -1, 6, -1, -1, -1 }, // 3121
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, -1, -1, 7, -1, -1, -1 }, // 4121
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, -1, -1, 5, -1, -1, -1 }, // 1221
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, -1, -1, 6, -1, -1, -1 }, // 2221
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, -1, -1, 7, -1, -1, -1 }, // 3221
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, -1, -1, 8, -1, -1, -1 }, // 4221
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, -1, -1, 6, -1, -1, -1 }, // 1321
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, -1, -1, 7, -1, -1, -1 }, // 2321
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, -1, -1, 8, -1, -1, -1 }, // 3321
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, -1, -1, 9, -1, -1, -1 }, // 4321
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, -1, -1, 7, -1, -1, -1 }, // 1421
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, -1, -1, 8, -1, -1, -1 }, // 2421
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, -1, -1, 9, -1, -1, -1 }, // 3421
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, 10, -1, -1, -1 }, // 4421
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, -1, 5, -1, -1, -1 }, // 1131
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, -1, 6, -1, -1, -1 }, // 2131
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, -1, 7, -1, -1, -1 }, // 3131
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, -1, 8, -1, -1, -1 }, // 4131
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, -1, 6, -1, -1, -1 }, // 1231
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, -1, 7, -1, -1, -1 }, // 2231
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, -1, 8, -1, -1, -1 }, // 3231
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, -1, 9, -1, -1, -1 }, // 4231
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, -1, 7, -1, -1, -1 }, // 1331
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, -1, 8, -1, -1, -1 }, // 2331
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, -1, 9, -1, -1, -1 }, // 3331
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, -1, 10, -1, -1, -1 }, // 4331
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, -1, 8, -1, -1, -1 }, // 1431
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, -1, 9, -1, -1, -1 }, // 2431
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, -1, 10, -1, -1, -1 }, // 3431
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, -1, 11, -1, -1, -1 }, // 4431
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, 5, 6, -1, -1, -1 }, // 1141
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, 6, 7, -1, -1, -1 }, // 2141
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, 7, 8, -1, -1, -1 }, // 3141
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, 8, 9, -1, -1, -1 }, // 4141
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, 6, 7, -1, -1, -1 }, // 1241
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, 7, 8, -1, -1, -1 }, // 2241
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, 8, 9, -1, -1, -1 }, // 3241
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, 9, 10, -1, -1, -1 }, // 4241
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, 7, 8, -1, -1, -1 }, // 1341
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, 8, 9, -1, -1, -1 }, // 2341
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, 9, 10, -1, -1, -1 }, // 3341
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, 10, 11, -1, -1, -1 }, // 4341
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, -1 }, // 1441
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, 9, 10, -1, -1, -1 }, // 2441
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, 10, 11, -1, -1, -1 }, // 3441
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, -1, -1, -1 }, // 4441
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, -1, -1, -1, 3, 4, -1, -1 }, // 1112
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, -1, -1, -1, 4, 5, -1, -1 }, // 2112
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, -1, -1, -1, 5, 6, -1, -1 }, // 3112
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, -1, -1, -1, 6, 7, -1, -1 }, // 4112
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, -1, -1, -1, 4, 5, -1, -1 }, // 1212
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, -1, -1, -1, 5, 6, -1, -1 }, // 2212
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, -1, -1, -1, 6, 7, -1, -1 }, // 3212
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, -1, -1, -1, 7, 8, -1, -1 }, // 4212
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, -1, -1, -1, 5, 6, -1, -1 }, // 1312
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, -1, -1, -1, 6, 7, -1, -1 }, // 2312
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, -1, -1, -1, 7, 8, -1, -1 }, // 3312
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, -1, -1, -1, 8, 9, -1, -1 }, // 4312
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, -1, -1, -1, 6, 7, -1, -1 }, // 1412
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, -1, -1, -1, 7, 8, -1, -1 }, // 2412
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, -1, -1, -1, 8, 9, -1, -1 }, // 3412
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, -1, -1, -1, 9, 10, -1, -1 }, // 4412
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, -1, -1, 4, 5, -1, -1 }, // 1122
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, -1, -1, 5, 6, -1, -1 }, // 2122
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, -1, -1, 6, 7, -1, -1 }, // 3122
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, -1, -1, 7, 8, -1, -1 }, // 4122
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, -1, -1, 5, 6, -1, -1 }, // 1222
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, -1, -1, 6, 7, -1, -1 }, // 2222
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, -1, -1, 7, 8, -1, -1 }, // 3222
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, -1, -1, 8, 9, -1, -1 }, // 4222
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, -1, -1, 6, 7, -1, -1 }, // 1322
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, -1, -1, 7, 8, -1, -1 }, // 2322
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, -1, -1, 8, 9, -1, -1 }, // 3322
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, -1, -1, 9, 10, -1, -1 }, // 4322
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, -1, -1, 7, 8, -1, -1 }, // 1422
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, -1, -1, 8, 9, -1, -1 }, // 2422
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, -1, -1, 9, 10, -1, -1 }, // 3422
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, 10, 11, -1, -1 }, // 4422
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, -1, 5, 6, -1, -1 }, // 1132
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, -1, 6, 7, -1, -1 }, // 2132
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, -1, 7, 8, -1, -1 }, // 3132
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, -1, 8, 9, -1, -1 }, // 4132
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, -1, 6, 7, -1, -1 }, // 1232
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, -1, 7, 8, -1, -1 }, // 2232
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, -1, 8, 9, -1, -1 }, // 3232
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, -1, 9, 10, -1, -1 }, // 4232
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, -1, 7, 8, -1, -1 }, // 1332
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, -1, 8, 9, -1, -1 }, // 2332
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, -1, 9, 10, -1, -1 }, // 3332
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, -1, 10, 11, -1, -1 }, // 4332
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, -1, 8, 9, -1, -1 }, // 1432
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, -1, 9, 10, -1, -1 }, // 2432
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, -1, 10, 11, -1, -1 }, // 3432
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, -1, 11, 12, -1, -1 }, // 4432
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, 5, 6, 7, -1, -1 }, // 1142
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, 6, 7, 8, -1, -1 }, // 2142
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, 7, 8, 9, -1, -1 }, // 3142
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, 8, 9, 10, -1, -1 }, // 4142
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, 6, 7, 8, -1, -1 }, // 1242
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, 7, 8, 9, -1, -1 }, // 2242
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, 8, 9, 10, -1, -1 }, // 3242
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, 9, 10, 11, -1, -1 }, // 4242
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, 7, 8, 9, -1, -1 }, // 1342
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, 8, 9, 10, -1, -1 }, // 2342
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, 9, 10, 11, -1, -1 }, // 3342
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, 10, 11, 12, -1, -1 }, // 4342
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, -1, -1 }, // 1442
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, -1, -1 }, // 2442
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, -1, -1 }, // 3442
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, -1, -1 }, // 4442
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, -1, -1, -1, 3, 4, 5, -1 }, // 1113
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, -1, -1, -1, 4, 5, 6, -1 }, // 2113
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, -1, -1, -1, 5, 6, 7, -1 }, // 3113
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, -1, -1, -1, 6, 7, 8, -1 }, // 4113
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, -1, -1, -1, 4, 5, 6, -1 }, // 1213
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, -1, -1, -1, 5, 6, 7, -1 }, // 2213
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, -1, -1, -1, 6, 7, 8, -1 }, // 3213
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, -1, -1, -1, 7, 8, 9, -1 }, // 4213
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, -1, -1, -1, 5, 6, 7, -1 }, // 1313
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, -1, -1, -1, 6, 7, 8, -1 }, // 2313
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, -1, -1, -1, 7, 8, 9, -1 }, // 3313
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, -1, -1, -1, 8, 9, 10, -1 }, // 4313
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, -1, -1, -1, 6, 7, 8, -1 }, // 1413
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, -1, -1, -1, 7, 8, 9, -1 }, // 2413
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, -1, -1, -1, 8, 9, 10, -1 }, // 3413
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, -1, -1, -1, 9, 10, 11, -1 }, // 4413
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, -1, -1, 4, 5, 6, -1 }, // 1123
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, -1, -1, 5, 6, 7, -1 }, // 2123
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, -1, -1, 6, 7, 8, -1 }, // 3123
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, -1, -1, 7, 8, 9, -1 }, // 4123
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, -1, -1, 5, 6, 7, -1 }, // 1223
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, -1, -1, 6, 7, 8, -1 }, // 2223
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, -1, -1, 7, 8, 9, -1 }, // 3223
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, -1, -1, 8, 9, 10, -1 }, // 4223
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, -1, -1, 6, 7, 8, -1 }, // 1323
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, -1, -1, 7, 8, 9, -1 }, // 2323
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, -1, -1, 8, 9, 10, -1 }, // 3323
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, -1, -1, 9, 10, 11, -1 }, // 4323
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, -1, -1, 7, 8, 9, -1 }, // 1423
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, -1, -1, 8, 9, 10, -1 }, // 2423
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, -1, -1, 9, 10, 11, -1 }, // 3423
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, 10, 11, 12, -1 }, // 4423
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, -1, 5, 6, 7, -1 }, // 1133
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, -1, 6, 7, 8, -1 }, // 2133
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, -1, 7, 8, 9, -1 }, // 3133
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, -1, 8, 9, 10, -1 }, // 4133
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, -1, 6, 7, 8, -1 }, // 1233
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, -1, 7, 8, 9, -1 }, // 2233
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, -1, 8, 9, 10, -1 }, // 3233
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, -1, 9, 10, 11, -1 }, // 4233
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, -1, 7, 8, 9, -1 }, // 1333
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, -1, 8, 9, 10, -1 }, // 2333
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, -1, 9, 10, 11, -1 }, // 3333
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, -1, 10, 11, 12, -1 }, // 4333
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, -1, 8, 9, 10, -1 }, // 1433
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, -1, 9, 10, 11, -1 }, // 2433
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, -1, 10, 11, 12, -1 }, // 3433
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, -1, 11, 12, 13, -1 }, // 4433
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, 5, 6, 7, 8, -1 }, // 1143
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, 6, 7, 8, 9, -1 }, // 2143
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, 7, 8, 9, 10, -1 }, // 3143
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, 8, 9, 10, 11, -1 }, // 4143
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, 6, 7, 8, 9, -1 }, // 1243
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, 7, 8, 9, 10, -1 }, // 2243
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, 8, 9, 10, 11, -1 }, // 3243
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, 9, 10, 11, 12, -1 }, // 4243
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, 7, 8, 9, 10, -1 }, // 1343
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, 8, 9, 10, 11, -1 }, // 2343
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, 9, 10, 11, 12, -1 }, // 3343
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, 10, 11, 12, 13, -1 }, // 4343
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, -1 }, // 1443
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, -1 }, // 2443
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, -1 }, // 3443
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, -1 }, // 4443
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, -1, -1, -1, 3, 4, 5, 6 }, // 1114
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, -1, -1, -1, 4, 5, 6, 7 }, // 2114
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, -1, -1, -1, 5, 6, 7, 8 }, // 3114
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, -1, -1, -1, 6, 7, 8, 9 }, // 4114
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, -1, -1, -1, 4, 5, 6, 7 }, // 1214
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, -1, -1, -1, 5, 6, 7, 8 }, // 2214
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, -1, -1, -1, 6, 7, 8, 9 }, // 3214
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, -1, -1, -1, 7, 8, 9, 10 }, // 4214
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, -1, -1, -1, 5, 6, 7, 8 }, // 1314
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, -1, -1, -1, 6, 7, 8, 9 }, // 2314
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, -1, -1, -1, 7, 8, 9, 10 }, // 3314
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, -1, -1, -1, 8, 9, 10, 11 }, // 4314
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, -1, -1, -1, 6, 7, 8, 9 }, // 1414
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, -1, -1, -1, 7, 8, 9, 10 }, // 2414
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, -1, -1, -1, 8, 9, 10, 11 }, // 3414
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, -1, -1, -1, 9, 10, 11, 12 }, // 4414
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, -1, -1, 4, 5, 6, 7 }, // 1124
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, -1, -1, 5, 6, 7, 8 }, // 2124
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, -1, -1, 6, 7, 8, 9 }, // 3124
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, -1, -1, 7, 8, 9, 10 }, // 4124
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, -1, -1, 5, 6, 7, 8 }, // 1224
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, -1, -1, 6, 7, 8, 9 }, // 2224
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, -1, -1, 7, 8, 9, 10 }, // 3224
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, -1, -1, 8, 9, 10, 11 }, // 4224
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, -1, -1, 6, 7, 8, 9 }, // 1324
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, -1, -1, 7, 8, 9, 10 }, // 2324
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, -1, -1, 8, 9, 10, 11 }, // 3324
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, -1, -1, 9, 10, 11, 12 }, // 4324
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, -1, -1, 7, 8, 9, 10 }, // 1424
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, -1, -1, 8, 9, 10, 11 }, // 2424
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, -1, -1, 9, 10, 11, 12 }, // 3424
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, 10, 11, 12, 13 }, // 4424
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, -1, 5, 6, 7, 8 }, // 1134
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, -1, 6, 7, 8, 9 }, // 2134
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, -1, 7, 8, 9, 10 }, // 3134
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, -1, 8, 9, 10, 11 }, // 4134
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, -1, 6, 7, 8, 9 }, // 1234
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, -1, 7, 8, 9, 10 }, // 2234
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, -1, 8, 9, 10, 11 }, // 3234
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, -1, 9, 10, 11, 12 }, // 4234
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, -1, 7, 8, 9, 10 }, // 1334
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, -1, 8, 9, 10, 11 }, // 2334
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, -1, 9, 10, 11, 12 }, // 3334
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, -1, 10, 11, 12, 13 }, // 4334
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, -1, 8, 9, 10, 11 }, // 1434
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, -1, 9, 10, 11, 12 }, // 2434
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, -1, 10, 11, 12, 13 }, // 3434
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, -1, 11, 12, 13, 14 }, // 4434
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, 5, 6, 7, 8, 9 }, // 1144
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, 6, 7, 8, 9, 10 }, // 2144
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, 7, 8, 9, 10, 11 }, // 3144
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, 8, 9, 10, 11, 12 }, // 4144
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, 6, 7, 8, 9, 10 }, // 1244
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, 7, 8, 9, 10, 11 }, // 2244
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, 8, 9, 10, 11, 12 }, // 3244
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, 9, 10, 11, 12, 13 }, // 4244
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, 7, 8, 9, 10, 11 }, // 1344
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, 8, 9, 10, 11, 12 }, // 2344
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, 9, 10, 11, 12, 13 }, // 3344
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, 10, 11, 12, 13, 14 }, // 4344
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 }, // 1444
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13 }, // 2444
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 }, // 3444
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 } // 4444
};
static uint8_t _encode_data(uint32_t val, uint8_t *__restrict__ *dataPtrPtr) {
uint8_t *dataPtr = *dataPtrPtr;
uint8_t code;
if (val < (1 << 8)) { // 1 byte
*dataPtr = (uint8_t)(val);
*dataPtrPtr += 1;
code = 0;
} else if (val < (1 << 16)) { // 2 bytes
*(uint16_t *) dataPtr = (uint16_t)(val);
*dataPtrPtr += 2;
code = 1;
} else if (val < (1 << 24)) { // 3 bytes
*(uint16_t *) dataPtr = (uint16_t)(val);
*(dataPtr + 2) = (uint8_t)(val >> 16);
*dataPtrPtr += 3;
code = 2;
} else { // 4 bytes
*(uint32_t *) dataPtr = val;
*dataPtrPtr += 4;
code = 3;
}
return code;
}
static uint8_t *svb_encode_scalar(const uint32_t *in,
uint8_t *__restrict__ keyPtr, uint8_t *__restrict__ dataPtr,
uint32_t count) {
if (count == 0)
return dataPtr; // exit immediately if no data
uint8_t shift = 0; // cycles 0, 2, 4, 6, 0, 2, 4, 6, ...
uint8_t key = 0;
for (uint32_t c = 0; c < count; c++) {
if (shift == 8) {
shift = 0;
*keyPtr++ = key;
key = 0;
}
uint32_t val = in[c];
uint8_t code = _encode_data(val, &dataPtr);
key |= code << shift;
shift += 2;
}
*keyPtr = key; // write last key (no increment needed)
return dataPtr; // pointer to first unused data byte
}
// Encode an array of a given length read from in to bout in streamvbyte format.
// Returns the number of bytes written.
size_t streamvbyte_encode(const uint32_t *in, uint32_t count, uint8_t *out) {
uint8_t *keyPtr = out;
uint32_t keyLen = (count + 3) / 4; // 2-bits rounded to full byte
uint8_t *dataPtr = keyPtr + keyLen; // variable byte data after all keys
return svb_encode_scalar(in, keyPtr, dataPtr, count) - out;
}
static inline __m128i _decode_avx(uint32_t key,
const uint8_t *__restrict__ *dataPtrPtr) {
uint8_t len = lengthTable[key];
__m128i Data = _mm_loadu_si128((__m128i *) *dataPtrPtr);
__m128i Shuf = *(__m128i *) &shuffleTable[key];
Data = _mm_shuffle_epi8(Data, Shuf);
*dataPtrPtr += len;
return Data;
}
static inline void _write_avx(uint32_t *out, __m128i Vec) {
_mm_storeu_si128((__m128i *) out, Vec);
}
static inline uint32_t _decode_data(const uint8_t **dataPtrPtr, uint8_t code) {
const uint8_t *dataPtr = *dataPtrPtr;
uint32_t val;
if (code == 0) { // 1 byte
val = (uint32_t) * dataPtr;
dataPtr += 1;
} else if (code == 1) { // 2 bytes
val = (uint32_t) * (uint16_t *) dataPtr;
dataPtr += 2;
} else if (code == 2) { // 3 bytes
val = (uint32_t) * (uint16_t *) dataPtr;
val |= *(dataPtr + 2) << 16;
dataPtr += 3;
} else { // code == 3
val = *(uint32_t *) dataPtr; // 4 bytes
dataPtr += 4;
}
*dataPtrPtr = dataPtr;
return val;
}
static const uint8_t *svb_decode_scalar(uint32_t *outPtr, const uint8_t *keyPtr,
const uint8_t *dataPtr, uint32_t count) {
if (count == 0)
return dataPtr; // no reads or writes if no data
uint8_t shift = 0;
uint32_t key = *keyPtr++;
for (uint32_t c = 0; c < count; c++) {
if (shift == 8) {
shift = 0;
key = *keyPtr++;
}
uint32_t val = _decode_data(&dataPtr, (key >> shift) & 0x3);
*outPtr++ = val;
shift += 2;
}
return dataPtr; // pointer to first unused byte after end
}
const uint8_t *svb_decode_avx_simple(uint32_t *out,
const uint8_t *__restrict__ keyPtr, const uint8_t *__restrict__ dataPtr,
uint64_t count) {
uint64_t keybytes = count / 4; // number of key bytes
__m128i Data;
if (keybytes >= 8) {
int64_t Offset = -(int64_t) keybytes / 8 + 1;
const uint64_t *keyPtr64 = (const uint64_t *) keyPtr - Offset;
uint64_t nextkeys = keyPtr64[Offset];
for (; Offset != 0; ++Offset) {
uint64_t keys = nextkeys;
nextkeys = keyPtr64[Offset + 1];
Data = _decode_avx((keys & 0xFF), &dataPtr);
_write_avx(out, Data);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
_write_avx(out + 4, Data);
keys >>= 16;
Data = _decode_avx((keys & 0xFF), &dataPtr);
_write_avx(out + 8, Data);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
_write_avx(out + 12, Data);
keys >>= 16;
Data = _decode_avx((keys & 0xFF), &dataPtr);
_write_avx(out + 16, Data);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
_write_avx(out + 20, Data);
keys >>= 16;
Data = _decode_avx((keys & 0xFF), &dataPtr);
_write_avx(out + 24, Data);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
_write_avx(out + 28, Data);
out += 32;
}
{
uint64_t keys = nextkeys;
Data = _decode_avx((keys & 0xFF), &dataPtr);
_write_avx(out, Data);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
_write_avx(out + 4, Data);
keys >>= 16;
Data = _decode_avx((keys & 0xFF), &dataPtr);
_write_avx(out + 8, Data);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
_write_avx(out + 12, Data);
keys >>= 16;
Data = _decode_avx((keys & 0xFF), &dataPtr);
_write_avx(out + 16, Data);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
_write_avx(out + 20, Data);
keys >>= 16;
Data = _decode_avx((keys & 0xFF), &dataPtr);
_write_avx(out + 24, Data);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
_write_avx(out + 28, Data);
out += 32;
}
}
uint64_t consumedkeys = keybytes - (keybytes & 7);
return svb_decode_scalar(out, keyPtr + consumedkeys, dataPtr, count & 31);
}
// Read count 32-bit integers in maskedvbyte format from in, storing the result in out. Returns the number of bytes read.
size_t streamvbyte_decode(const uint8_t* in, uint32_t* out, uint32_t count) {
if (count == 0)
return 0;
const uint8_t *keyPtr = in; // full list of keys is next
uint32_t keyLen = ((count + 3) / 4); // 2-bits per key (rounded up)
const uint8_t *dataPtr = keyPtr + keyLen; // data starts at end of keys
return svb_decode_avx_simple(out, keyPtr, dataPtr, count) - in;
}

575
cpp/streamvbyte/src/streamvbytedelta.c Normal file
View File

@@ -0,0 +1,575 @@
#include "streamvbyte.h"
#if defined(_MSC_VER)
/* Microsoft C/C++-compatible compiler */
#include <intrin.h>
#elif defined(__GNUC__) && (defined(__x86_64__) || defined(__i386__))
/* GCC-compatible compiler, targeting x86/x86-64 */
#include <x86intrin.h>
#elif defined(__GNUC__) && defined(__ARM_NEON__)
/* GCC-compatible compiler, targeting ARM with NEON */
#include <arm_neon.h>
#elif defined(__GNUC__) && defined(__IWMMXT__)
/* GCC-compatible compiler, targeting ARM with WMMX */
#include <mmintrin.h>
#elif (defined(__GNUC__) || defined(__xlC__)) && (defined(__VEC__) || defined(__ALTIVEC__))
/* XLC or GCC-compatible compiler, targeting PowerPC with VMX/VSX */
#include <altivec.h>
#elif defined(__GNUC__) && defined(__SPE__)
/* GCC-compatible compiler, targeting PowerPC with SPE */
#include <spe.h>
#endif
static uint8_t lengthTable[256] = { 4, 5, 6, 7, 5, 6, 7, 8, 6, 7, 8, 9, 7, 8, 9,
10, 5, 6, 7, 8, 6, 7, 8, 9, 7, 8, 9, 10, 8, 9, 10, 11, 6, 7, 8, 9, 7, 8,
9, 10, 8, 9, 10, 11, 9, 10, 11, 12, 7, 8, 9, 10, 8, 9, 10, 11, 9, 10,
11, 12, 10, 11, 12, 13, 5, 6, 7, 8, 6, 7, 8, 9, 7, 8, 9, 10, 8, 9, 10,
11, 6, 7, 8, 9, 7, 8, 9, 10, 8, 9, 10, 11, 9, 10, 11, 12, 7, 8, 9, 10,
8, 9, 10, 11, 9, 10, 11, 12, 10, 11, 12, 13, 8, 9, 10, 11, 9, 10, 11,
12, 10, 11, 12, 13, 11, 12, 13, 14, 6, 7, 8, 9, 7, 8, 9, 10, 8, 9, 10,
11, 9, 10, 11, 12, 7, 8, 9, 10, 8, 9, 10, 11, 9, 10, 11, 12, 10, 11, 12,
13, 8, 9, 10, 11, 9, 10, 11, 12, 10, 11, 12, 13, 11, 12, 13, 14, 9, 10,
11, 12, 10, 11, 12, 13, 11, 12, 13, 14, 12, 13, 14, 15, 7, 8, 9, 10, 8,
9, 10, 11, 9, 10, 11, 12, 10, 11, 12, 13, 8, 9, 10, 11, 9, 10, 11, 12,
10, 11, 12, 13, 11, 12, 13, 14, 9, 10, 11, 12, 10, 11, 12, 13, 11, 12,
13, 14, 12, 13, 14, 15, 10, 11, 12, 13, 11, 12, 13, 14, 12, 13, 14, 15,
13, 14, 15, 16 };
static uint8_t shuffleTable[256][16] = { { 0, -1, -1, -1, 1, -1, -1, -1, 2, -1,
-1, -1, 3, -1, -1, -1 }, // 1111
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, -1, -1, -1, 4, -1, -1, -1 }, // 2111
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, -1, -1, -1, 5, -1, -1, -1 }, // 3111
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, -1, -1, -1, 6, -1, -1, -1 }, // 4111
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, -1, -1, -1, 4, -1, -1, -1 }, // 1211
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, -1, -1, -1, 5, -1, -1, -1 }, // 2211
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, -1, -1, -1, 6, -1, -1, -1 }, // 3211
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, -1, -1, -1, 7, -1, -1, -1 }, // 4211
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, -1, -1, -1, 5, -1, -1, -1 }, // 1311
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, -1, -1, -1, 6, -1, -1, -1 }, // 2311
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, -1, -1, -1, 7, -1, -1, -1 }, // 3311
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, -1, -1, -1, 8, -1, -1, -1 }, // 4311
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, -1, -1, -1, 6, -1, -1, -1 }, // 1411
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, -1, -1, -1, 7, -1, -1, -1 }, // 2411
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, -1, -1, -1, 8, -1, -1, -1 }, // 3411
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, -1, -1, -1, 9, -1, -1, -1 }, // 4411
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, -1, -1, 4, -1, -1, -1 }, // 1121
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, -1, -1, 5, -1, -1, -1 }, // 2121
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, -1, -1, 6, -1, -1, -1 }, // 3121
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, -1, -1, 7, -1, -1, -1 }, // 4121
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, -1, -1, 5, -1, -1, -1 }, // 1221
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, -1, -1, 6, -1, -1, -1 }, // 2221
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, -1, -1, 7, -1, -1, -1 }, // 3221
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, -1, -1, 8, -1, -1, -1 }, // 4221
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, -1, -1, 6, -1, -1, -1 }, // 1321
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, -1, -1, 7, -1, -1, -1 }, // 2321
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, -1, -1, 8, -1, -1, -1 }, // 3321
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, -1, -1, 9, -1, -1, -1 }, // 4321
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, -1, -1, 7, -1, -1, -1 }, // 1421
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, -1, -1, 8, -1, -1, -1 }, // 2421
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, -1, -1, 9, -1, -1, -1 }, // 3421
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, 10, -1, -1, -1 }, // 4421
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, -1, 5, -1, -1, -1 }, // 1131
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, -1, 6, -1, -1, -1 }, // 2131
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, -1, 7, -1, -1, -1 }, // 3131
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, -1, 8, -1, -1, -1 }, // 4131
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, -1, 6, -1, -1, -1 }, // 1231
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, -1, 7, -1, -1, -1 }, // 2231
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, -1, 8, -1, -1, -1 }, // 3231
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, -1, 9, -1, -1, -1 }, // 4231
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, -1, 7, -1, -1, -1 }, // 1331
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, -1, 8, -1, -1, -1 }, // 2331
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, -1, 9, -1, -1, -1 }, // 3331
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, -1, 10, -1, -1, -1 }, // 4331
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, -1, 8, -1, -1, -1 }, // 1431
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, -1, 9, -1, -1, -1 }, // 2431
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, -1, 10, -1, -1, -1 }, // 3431
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, -1, 11, -1, -1, -1 }, // 4431
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, 5, 6, -1, -1, -1 }, // 1141
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, 6, 7, -1, -1, -1 }, // 2141
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, 7, 8, -1, -1, -1 }, // 3141
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, 8, 9, -1, -1, -1 }, // 4141
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, 6, 7, -1, -1, -1 }, // 1241
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, 7, 8, -1, -1, -1 }, // 2241
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, 8, 9, -1, -1, -1 }, // 3241
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, 9, 10, -1, -1, -1 }, // 4241
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, 7, 8, -1, -1, -1 }, // 1341
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, 8, 9, -1, -1, -1 }, // 2341
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, 9, 10, -1, -1, -1 }, // 3341
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, 10, 11, -1, -1, -1 }, // 4341
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, -1 }, // 1441
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, 9, 10, -1, -1, -1 }, // 2441
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, 10, 11, -1, -1, -1 }, // 3441
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, -1, -1, -1 }, // 4441
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, -1, -1, -1, 3, 4, -1, -1 }, // 1112
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, -1, -1, -1, 4, 5, -1, -1 }, // 2112
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, -1, -1, -1, 5, 6, -1, -1 }, // 3112
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, -1, -1, -1, 6, 7, -1, -1 }, // 4112
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, -1, -1, -1, 4, 5, -1, -1 }, // 1212
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, -1, -1, -1, 5, 6, -1, -1 }, // 2212
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, -1, -1, -1, 6, 7, -1, -1 }, // 3212
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, -1, -1, -1, 7, 8, -1, -1 }, // 4212
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, -1, -1, -1, 5, 6, -1, -1 }, // 1312
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, -1, -1, -1, 6, 7, -1, -1 }, // 2312
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, -1, -1, -1, 7, 8, -1, -1 }, // 3312
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, -1, -1, -1, 8, 9, -1, -1 }, // 4312
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, -1, -1, -1, 6, 7, -1, -1 }, // 1412
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, -1, -1, -1, 7, 8, -1, -1 }, // 2412
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, -1, -1, -1, 8, 9, -1, -1 }, // 3412
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, -1, -1, -1, 9, 10, -1, -1 }, // 4412
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, -1, -1, 4, 5, -1, -1 }, // 1122
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, -1, -1, 5, 6, -1, -1 }, // 2122
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, -1, -1, 6, 7, -1, -1 }, // 3122
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, -1, -1, 7, 8, -1, -1 }, // 4122
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, -1, -1, 5, 6, -1, -1 }, // 1222
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, -1, -1, 6, 7, -1, -1 }, // 2222
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, -1, -1, 7, 8, -1, -1 }, // 3222
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, -1, -1, 8, 9, -1, -1 }, // 4222
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, -1, -1, 6, 7, -1, -1 }, // 1322
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, -1, -1, 7, 8, -1, -1 }, // 2322
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, -1, -1, 8, 9, -1, -1 }, // 3322
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, -1, -1, 9, 10, -1, -1 }, // 4322
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, -1, -1, 7, 8, -1, -1 }, // 1422
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, -1, -1, 8, 9, -1, -1 }, // 2422
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, -1, -1, 9, 10, -1, -1 }, // 3422
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, 10, 11, -1, -1 }, // 4422
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, -1, 5, 6, -1, -1 }, // 1132
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, -1, 6, 7, -1, -1 }, // 2132
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, -1, 7, 8, -1, -1 }, // 3132
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, -1, 8, 9, -1, -1 }, // 4132
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, -1, 6, 7, -1, -1 }, // 1232
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, -1, 7, 8, -1, -1 }, // 2232
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, -1, 8, 9, -1, -1 }, // 3232
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, -1, 9, 10, -1, -1 }, // 4232
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, -1, 7, 8, -1, -1 }, // 1332
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, -1, 8, 9, -1, -1 }, // 2332
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, -1, 9, 10, -1, -1 }, // 3332
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, -1, 10, 11, -1, -1 }, // 4332
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, -1, 8, 9, -1, -1 }, // 1432
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, -1, 9, 10, -1, -1 }, // 2432
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, -1, 10, 11, -1, -1 }, // 3432
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, -1, 11, 12, -1, -1 }, // 4432
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, 5, 6, 7, -1, -1 }, // 1142
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, 6, 7, 8, -1, -1 }, // 2142
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, 7, 8, 9, -1, -1 }, // 3142
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, 8, 9, 10, -1, -1 }, // 4142
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, 6, 7, 8, -1, -1 }, // 1242
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, 7, 8, 9, -1, -1 }, // 2242
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, 8, 9, 10, -1, -1 }, // 3242
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, 9, 10, 11, -1, -1 }, // 4242
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, 7, 8, 9, -1, -1 }, // 1342
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, 8, 9, 10, -1, -1 }, // 2342
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, 9, 10, 11, -1, -1 }, // 3342
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, 10, 11, 12, -1, -1 }, // 4342
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, -1, -1 }, // 1442
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, -1, -1 }, // 2442
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, -1, -1 }, // 3442
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, -1, -1 }, // 4442
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, -1, -1, -1, 3, 4, 5, -1 }, // 1113
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, -1, -1, -1, 4, 5, 6, -1 }, // 2113
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, -1, -1, -1, 5, 6, 7, -1 }, // 3113
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, -1, -1, -1, 6, 7, 8, -1 }, // 4113
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, -1, -1, -1, 4, 5, 6, -1 }, // 1213
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, -1, -1, -1, 5, 6, 7, -1 }, // 2213
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, -1, -1, -1, 6, 7, 8, -1 }, // 3213
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, -1, -1, -1, 7, 8, 9, -1 }, // 4213
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, -1, -1, -1, 5, 6, 7, -1 }, // 1313
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, -1, -1, -1, 6, 7, 8, -1 }, // 2313
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, -1, -1, -1, 7, 8, 9, -1 }, // 3313
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, -1, -1, -1, 8, 9, 10, -1 }, // 4313
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, -1, -1, -1, 6, 7, 8, -1 }, // 1413
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, -1, -1, -1, 7, 8, 9, -1 }, // 2413
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, -1, -1, -1, 8, 9, 10, -1 }, // 3413
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, -1, -1, -1, 9, 10, 11, -1 }, // 4413
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, -1, -1, 4, 5, 6, -1 }, // 1123
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, -1, -1, 5, 6, 7, -1 }, // 2123
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, -1, -1, 6, 7, 8, -1 }, // 3123
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, -1, -1, 7, 8, 9, -1 }, // 4123
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, -1, -1, 5, 6, 7, -1 }, // 1223
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, -1, -1, 6, 7, 8, -1 }, // 2223
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, -1, -1, 7, 8, 9, -1 }, // 3223
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, -1, -1, 8, 9, 10, -1 }, // 4223
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, -1, -1, 6, 7, 8, -1 }, // 1323
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, -1, -1, 7, 8, 9, -1 }, // 2323
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, -1, -1, 8, 9, 10, -1 }, // 3323
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, -1, -1, 9, 10, 11, -1 }, // 4323
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, -1, -1, 7, 8, 9, -1 }, // 1423
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, -1, -1, 8, 9, 10, -1 }, // 2423
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, -1, -1, 9, 10, 11, -1 }, // 3423
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, 10, 11, 12, -1 }, // 4423
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, -1, 5, 6, 7, -1 }, // 1133
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, -1, 6, 7, 8, -1 }, // 2133
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, -1, 7, 8, 9, -1 }, // 3133
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, -1, 8, 9, 10, -1 }, // 4133
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, -1, 6, 7, 8, -1 }, // 1233
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, -1, 7, 8, 9, -1 }, // 2233
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, -1, 8, 9, 10, -1 }, // 3233
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, -1, 9, 10, 11, -1 }, // 4233
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, -1, 7, 8, 9, -1 }, // 1333
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, -1, 8, 9, 10, -1 }, // 2333
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, -1, 9, 10, 11, -1 }, // 3333
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, -1, 10, 11, 12, -1 }, // 4333
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, -1, 8, 9, 10, -1 }, // 1433
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, -1, 9, 10, 11, -1 }, // 2433
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, -1, 10, 11, 12, -1 }, // 3433
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, -1, 11, 12, 13, -1 }, // 4433
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, 5, 6, 7, 8, -1 }, // 1143
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, 6, 7, 8, 9, -1 }, // 2143
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, 7, 8, 9, 10, -1 }, // 3143
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, 8, 9, 10, 11, -1 }, // 4143
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, 6, 7, 8, 9, -1 }, // 1243
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, 7, 8, 9, 10, -1 }, // 2243
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, 8, 9, 10, 11, -1 }, // 3243
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, 9, 10, 11, 12, -1 }, // 4243
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, 7, 8, 9, 10, -1 }, // 1343
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, 8, 9, 10, 11, -1 }, // 2343
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, 9, 10, 11, 12, -1 }, // 3343
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, 10, 11, 12, 13, -1 }, // 4343
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, -1 }, // 1443
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, -1 }, // 2443
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, -1 }, // 3443
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, -1 }, // 4443
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, -1, -1, -1, 3, 4, 5, 6 }, // 1114
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, -1, -1, -1, 4, 5, 6, 7 }, // 2114
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, -1, -1, -1, 5, 6, 7, 8 }, // 3114
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, -1, -1, -1, 6, 7, 8, 9 }, // 4114
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, -1, -1, -1, 4, 5, 6, 7 }, // 1214
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, -1, -1, -1, 5, 6, 7, 8 }, // 2214
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, -1, -1, -1, 6, 7, 8, 9 }, // 3214
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, -1, -1, -1, 7, 8, 9, 10 }, // 4214
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, -1, -1, -1, 5, 6, 7, 8 }, // 1314
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, -1, -1, -1, 6, 7, 8, 9 }, // 2314
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, -1, -1, -1, 7, 8, 9, 10 }, // 3314
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, -1, -1, -1, 8, 9, 10, 11 }, // 4314
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, -1, -1, -1, 6, 7, 8, 9 }, // 1414
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, -1, -1, -1, 7, 8, 9, 10 }, // 2414
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, -1, -1, -1, 8, 9, 10, 11 }, // 3414
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, -1, -1, -1, 9, 10, 11, 12 }, // 4414
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, -1, -1, 4, 5, 6, 7 }, // 1124
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, -1, -1, 5, 6, 7, 8 }, // 2124
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, -1, -1, 6, 7, 8, 9 }, // 3124
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, -1, -1, 7, 8, 9, 10 }, // 4124
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, -1, -1, 5, 6, 7, 8 }, // 1224
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, -1, -1, 6, 7, 8, 9 }, // 2224
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, -1, -1, 7, 8, 9, 10 }, // 3224
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, -1, -1, 8, 9, 10, 11 }, // 4224
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, -1, -1, 6, 7, 8, 9 }, // 1324
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, -1, -1, 7, 8, 9, 10 }, // 2324
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, -1, -1, 8, 9, 10, 11 }, // 3324
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, -1, -1, 9, 10, 11, 12 }, // 4324
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, -1, -1, 7, 8, 9, 10 }, // 1424
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, -1, -1, 8, 9, 10, 11 }, // 2424
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, -1, -1, 9, 10, 11, 12 }, // 3424
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, 10, 11, 12, 13 }, // 4424
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, -1, 5, 6, 7, 8 }, // 1134
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, -1, 6, 7, 8, 9 }, // 2134
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, -1, 7, 8, 9, 10 }, // 3134
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, -1, 8, 9, 10, 11 }, // 4134
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, -1, 6, 7, 8, 9 }, // 1234
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, -1, 7, 8, 9, 10 }, // 2234
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, -1, 8, 9, 10, 11 }, // 3234
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, -1, 9, 10, 11, 12 }, // 4234
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, -1, 7, 8, 9, 10 }, // 1334
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, -1, 8, 9, 10, 11 }, // 2334
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, -1, 9, 10, 11, 12 }, // 3334
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, -1, 10, 11, 12, 13 }, // 4334
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, -1, 8, 9, 10, 11 }, // 1434
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, -1, 9, 10, 11, 12 }, // 2434
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, -1, 10, 11, 12, 13 }, // 3434
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, -1, 11, 12, 13, 14 }, // 4434
{ 0, -1, -1, -1, 1, -1, -1, -1, 2, 3, 4, 5, 6, 7, 8, 9 }, // 1144
{ 0, 1, -1, -1, 2, -1, -1, -1, 3, 4, 5, 6, 7, 8, 9, 10 }, // 2144
{ 0, 1, 2, -1, 3, -1, -1, -1, 4, 5, 6, 7, 8, 9, 10, 11 }, // 3144
{ 0, 1, 2, 3, 4, -1, -1, -1, 5, 6, 7, 8, 9, 10, 11, 12 }, // 4144
{ 0, -1, -1, -1, 1, 2, -1, -1, 3, 4, 5, 6, 7, 8, 9, 10 }, // 1244
{ 0, 1, -1, -1, 2, 3, -1, -1, 4, 5, 6, 7, 8, 9, 10, 11 }, // 2244
{ 0, 1, 2, -1, 3, 4, -1, -1, 5, 6, 7, 8, 9, 10, 11, 12 }, // 3244
{ 0, 1, 2, 3, 4, 5, -1, -1, 6, 7, 8, 9, 10, 11, 12, 13 }, // 4244
{ 0, -1, -1, -1, 1, 2, 3, -1, 4, 5, 6, 7, 8, 9, 10, 11 }, // 1344
{ 0, 1, -1, -1, 2, 3, 4, -1, 5, 6, 7, 8, 9, 10, 11, 12 }, // 2344
{ 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, 9, 10, 11, 12, 13 }, // 3344
{ 0, 1, 2, 3, 4, 5, 6, -1, 7, 8, 9, 10, 11, 12, 13, 14 }, // 4344
{ 0, -1, -1, -1, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 }, // 1444
{ 0, 1, -1, -1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13 }, // 2444
{ 0, 1, 2, -1, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 }, // 3444
{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 } // 4444
};
static uint8_t _encode_data(uint32_t val, uint8_t *__restrict__ *dataPtrPtr) {
uint8_t *dataPtr = *dataPtrPtr;
uint8_t code;
if (val < (1 << 8)) { // 1 byte
*dataPtr = (uint8_t)(val);
*dataPtrPtr += 1;
code = 0;
} else if (val < (1 << 16)) { // 2 bytes
*(uint16_t *) dataPtr = (uint16_t)(val);
*dataPtrPtr += 2;
code = 1;
} else if (val < (1 << 24)) { // 3 bytes
*(uint16_t *) dataPtr = (uint16_t)(val);
*(dataPtr + 2) = (uint8_t)(val >> 16);
*dataPtrPtr += 3;
code = 2;
} else { // 4 bytes
*(uint32_t *) dataPtr = val;
*dataPtrPtr += 4;
code = 3;
}
return code;
}
static uint8_t *svb_encode_scalar_d1_init(const uint32_t *in,
uint8_t *__restrict__ keyPtr, uint8_t *__restrict__ dataPtr,
uint32_t count, uint32_t prev) {
if (count == 0)
return dataPtr; // exit immediately if no data
uint8_t shift = 0; // cycles 0, 2, 4, 6, 0, 2, 4, 6, ...
uint8_t key = 0;
for (uint32_t c = 0; c < count; c++) {
if (shift == 8) {
shift = 0;
*keyPtr++ = key;
key = 0;
}
uint32_t val = in[c] - prev;
prev = in[c];
uint8_t code = _encode_data(val, &dataPtr);
key |= code << shift;
shift += 2;
}
*keyPtr = key; // write last key (no increment needed)
return dataPtr; // pointer to first unused data byte
}
size_t streamvbyte_delta_encode(const uint32_t *in, uint32_t count, uint8_t *out,
uint32_t prev) {
uint8_t *keyPtr = out; // keys come immediately after 32-bit count
uint32_t keyLen = (count + 3) / 4; // 2-bits rounded to full byte
uint8_t *dataPtr = keyPtr + keyLen; // variable byte data after all keys
return svb_encode_scalar_d1_init(in, keyPtr, dataPtr, count, prev) - out;
}
static inline __m128i _decode_avx(uint32_t key, const uint8_t *__restrict__ *dataPtrPtr) {
uint8_t len = lengthTable[key];
__m128i Data = _mm_loadu_si128((__m128i *) *dataPtrPtr);
__m128i Shuf = *(__m128i *) &shuffleTable[key];
Data = _mm_shuffle_epi8(Data, Shuf);
*dataPtrPtr += len;
return Data;
}
#define BroadcastLastXMM 0xFF // bits 0-7 all set to choose highest element
static inline void _write_avx(uint32_t *out, __m128i Vec) {
_mm_storeu_si128((__m128i *) out, Vec);
}
static __m128i _write_avx_d1(uint32_t *out, __m128i Vec, __m128i Prev) {
__m128i Add = _mm_slli_si128(Vec, 4); // Cycle 1: [- A B C] (already done)
Prev = _mm_shuffle_epi32(Prev, BroadcastLastXMM); // Cycle 2: [P P P P]
Vec = _mm_add_epi32(Vec, Add); // Cycle 2: [A AB BC CD]
Add = _mm_slli_si128(Vec, 8); // Cycle 3: [- - A AB]
Vec = _mm_add_epi32(Vec, Prev); // Cycle 3: [PA PAB PBC PCD]
Vec = _mm_add_epi32(Vec, Add); // Cycle 4: [PA PAB PABC PABCD]
_write_avx(out, Vec);
return Vec;
}
#ifndef _MSC_VER
static __m128i High16To32 = {0xFFFF0B0AFFFF0908, 0xFFFF0F0EFFFF0D0C};
#else
static __m128i High16To32 = {8, 9, -1, -1, 10, 11, -1, -1,
12, 13, -1, -1, 14, 15, -1, -1};
#endif
static inline __m128i _write_16bit_avx_d1(uint32_t *out, __m128i Vec, __m128i Prev) {
// vec == [A B C D E F G H] (16 bit values)
__m128i Add = _mm_slli_si128(Vec, 2); // [- A B C D E F G]
Prev = _mm_shuffle_epi32(Prev, BroadcastLastXMM); // [P P P P] (32-bit)
Vec = _mm_add_epi32(Vec, Add); // [A AB BC CD DE FG GH]
Add = _mm_slli_si128(Vec, 4); // [- - A AB BC CD DE EF]
Vec = _mm_add_epi32(Vec, Add); // [A AB ABC ABCD BCDE CDEF DEFG EFGH]
__m128i V1 = _mm_cvtepu16_epi32(Vec); // [A AB ABC ABCD] (32-bit)
V1 = _mm_add_epi32(V1, Prev); // [PA PAB PABC PABCD] (32-bit)
__m128i V2 =
_mm_shuffle_epi8(Vec, High16To32); // [BCDE CDEF DEFG EFGH] (32-bit)
V2 = _mm_add_epi32(V1, V2); // [PABCDE PABCDEF PABCDEFG PABCDEFGH] (32-bit)
_write_avx(out, V1);
_write_avx(out + 4, V2);
return V2;
}
static inline uint32_t _decode_data(const uint8_t **dataPtrPtr, uint8_t code) {
const uint8_t *dataPtr = *dataPtrPtr;
uint32_t val;
if (code == 0) { // 1 byte
val = (uint32_t) * dataPtr;
dataPtr += 1;
} else if (code == 1) { // 2 bytes
val = (uint32_t) * (uint16_t *) dataPtr;
dataPtr += 2;
} else if (code == 2) { // 3 bytes
val = (uint32_t) * (uint16_t *) dataPtr;
val |= *(dataPtr + 2) << 16;
dataPtr += 3;
} else { // code == 3
val = *(uint32_t *) dataPtr; // 4 bytes
dataPtr += 4;
}
*dataPtrPtr = dataPtr;
return val;
}
const uint8_t *svb_decode_scalar_d1_init(uint32_t *outPtr, const uint8_t *keyPtr,
const uint8_t *dataPtr, uint32_t count,
uint32_t prev) {
if (count == 0)
return dataPtr; // no reads or writes if no data
uint8_t shift = 0;
uint32_t key = *keyPtr++;
for (uint32_t c = 0; c < count; c++) {
if (shift == 8) {
shift = 0;
key = *keyPtr++;
}
uint32_t val = _decode_data(&dataPtr, (key >> shift) & 0x3);
val += prev;
*outPtr++ = val;
prev = val;
shift += 2;
}
return dataPtr; // pointer to first unused byte after end
}
const uint8_t *svb_decode_avx_d1_init(uint32_t *out, const uint8_t *__restrict__ keyPtr,
const uint8_t *__restrict__ dataPtr, uint64_t count, uint32_t prev) {
uint64_t keybytes = count / 4; // number of key bytes
if (keybytes >= 8) {
__m128i Prev = _mm_set1_epi32(prev);
__m128i Data;
int64_t Offset = -(int64_t) keybytes / 8 + 1;
const uint64_t *keyPtr64 = (const uint64_t *) keyPtr - Offset;
uint64_t nextkeys = keyPtr64[Offset];
for (; Offset != 0; ++Offset) {
uint64_t keys = nextkeys;
nextkeys = keyPtr64[Offset + 1];
// faster 16-bit delta since we only have 8-bit values
if (!keys) { // 32 1-byte ints in a row
Data = _mm_cvtepu8_epi16(_mm_lddqu_si128((__m128i *) (dataPtr)));
Prev = _write_16bit_avx_d1(out, Data, Prev);
Data = _mm_cvtepu8_epi16(
_mm_lddqu_si128((__m128i *) (dataPtr + 8)));
Prev = _write_16bit_avx_d1(out + 8, Data, Prev);
Data = _mm_cvtepu8_epi16(
_mm_lddqu_si128((__m128i *) (dataPtr + 16)));
Prev = _write_16bit_avx_d1(out + 16, Data, Prev);
Data = _mm_cvtepu8_epi16(
_mm_lddqu_si128((__m128i *) (dataPtr + 24)));
Prev = _write_16bit_avx_d1(out + 24, Data, Prev);
out += 32;
dataPtr += 32;
continue;
}
Data = _decode_avx(keys & 0x00FF, &dataPtr);
Prev = _write_avx_d1(out, Data, Prev);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
Prev = _write_avx_d1(out + 4, Data, Prev);
keys >>= 16;
Data = _decode_avx((keys & 0x00FF), &dataPtr);
Prev = _write_avx_d1(out + 8, Data, Prev);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
Prev = _write_avx_d1(out + 12, Data, Prev);
keys >>= 16;
Data = _decode_avx((keys & 0x00FF), &dataPtr);
Prev = _write_avx_d1(out + 16, Data, Prev);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
Prev = _write_avx_d1(out + 20, Data, Prev);
keys >>= 16;
Data = _decode_avx((keys & 0x00FF), &dataPtr);
Prev = _write_avx_d1(out + 24, Data, Prev);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
Prev = _write_avx_d1(out + 28, Data, Prev);
out += 32;
}
{
uint64_t keys = nextkeys;
// faster 16-bit delta since we only have 8-bit values
if (!keys) { // 32 1-byte ints in a row
Data = _mm_cvtepu8_epi16(_mm_lddqu_si128((__m128i *) (dataPtr)));
Prev = _write_16bit_avx_d1(out, Data, Prev);
Data = _mm_cvtepu8_epi16(
_mm_lddqu_si128((__m128i *) (dataPtr + 8)));
Prev = _write_16bit_avx_d1(out + 8, Data, Prev);
Data = _mm_cvtepu8_epi16(
_mm_lddqu_si128((__m128i *) (dataPtr + 16)));
Prev = _write_16bit_avx_d1(out + 16, Data, Prev);
Data = _mm_cvtepu8_epi16(
_mm_loadl_epi64((__m128i *) (dataPtr + 24)));
Prev = _write_16bit_avx_d1(out + 24, Data, Prev);
out += 32;
dataPtr += 32;
} else {
Data = _decode_avx(keys & 0x00FF, &dataPtr);
Prev = _write_avx_d1(out, Data, Prev);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
Prev = _write_avx_d1(out + 4, Data, Prev);
keys >>= 16;
Data = _decode_avx((keys & 0x00FF), &dataPtr);
Prev = _write_avx_d1(out + 8, Data, Prev);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
Prev = _write_avx_d1(out + 12, Data, Prev);
keys >>= 16;
Data = _decode_avx((keys & 0x00FF), &dataPtr);
Prev = _write_avx_d1(out + 16, Data, Prev);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
Prev = _write_avx_d1(out + 20, Data, Prev);
keys >>= 16;
Data = _decode_avx((keys & 0x00FF), &dataPtr);
Prev = _write_avx_d1(out + 24, Data, Prev);
Data = _decode_avx((keys & 0xFF00) >> 8, &dataPtr);
Prev = _write_avx_d1(out + 28, Data, Prev);
out += 32;
}
}
prev = out[-1];
}
uint64_t consumedkeys = keybytes - (keybytes & 7);
return svb_decode_scalar_d1_init(out, keyPtr + consumedkeys, dataPtr,
count & 31, prev);
}
size_t streamvbyte_delta_decode(const uint8_t* in, uint32_t* out,
uint32_t count, uint32_t prev) {
uint32_t keyLen = ((count + 3) / 4); // 2-bits per key (rounded up)
const uint8_t *keyPtr = in;
const uint8_t *dataPtr = keyPtr + keyLen; // data starts at end of keys
return svb_decode_avx_d1_init(out, keyPtr, dataPtr, count, prev) - in;
}

73
cpp/streamvbyte/tests/unit.c Normal file
View File

@@ -0,0 +1,73 @@
#include <stdio.h>
#include <stdlib.h>
#include "streamvbyte.h"
#include "streamvbytedelta.h"
int main() {
int N = 4096;
uint32_t * datain = malloc(N * sizeof(uint32_t));
uint8_t * compressedbuffer = malloc(2 * N * sizeof(uint32_t));
uint32_t * recovdata = malloc(N * sizeof(uint32_t));
for (int length = 0; length <= N;) {
printf("length = %d \n", length);
for (uint32_t gap = 1; gap <= 387420489; gap *= 3) {
for (int k = 0; k < length; ++k)
datain[k] = gap;
size_t compsize = streamvbyte_encode(datain, length,
compressedbuffer);
size_t usedbytes = streamvbyte_decode(compressedbuffer, recovdata,
length);
if (compsize != usedbytes) {
printf(
"[streamvbyte_decode] code is buggy gap = %d, size mismatch %d %d \n",
(int) gap, (int) compsize, (int) usedbytes);
return -1;
}
for (int k = 0; k < length; ++k) {
if (recovdata[k] != datain[k]) {
printf("[streamvbyte_decode] code is buggy gap = %d\n",
(int) gap);
return -1;
}
}
}
printf("Delta \n");
for (size_t gap = 1; gap <= 531441; gap *= 3) {
for (int k = 0; k < length; ++k)
datain[k] = gap * k;
size_t compsize = streamvbyte_delta_encode(datain, length,
compressedbuffer, 0);
size_t usedbytes = streamvbyte_delta_decode(compressedbuffer,
recovdata, length, 0);
if (compsize != usedbytes) {
printf(
"[streamvbyte_delta_decode] code is buggy gap = %d, size mismatch %d %d \n",
(int) gap, (int) compsize, (int) usedbytes);
return -1;
}
for (int k = 0; k < length; ++k) {
if (recovdata[k] != datain[k]) {
printf(
"[streamvbyte_delta_decode] code is buggy gap = %d\n",
(int) gap);
return -1;
}
}
}
if (length < 128)
++length;
else {
length *= 2;
}
}
free(datain);
free(compressedbuffer);
free(recovdata);
printf("Code looks good.\n");
return 0;
}

217
examples/html/simple_search.html
View File

@@ -30,10 +30,12 @@
</div>
<div class="content"><div class='highlight'><pre><span class="hljs-keyword">extern</span> <span class="hljs-keyword">crate</span> rustc_serialize;
<span class="hljs-keyword">extern</span> <span class="hljs-keyword">crate</span> tantivy;
<div class="content"><div class='highlight'><pre><span class="hljs-keyword">extern</span> <span class="hljs-keyword">crate</span> tantivy;
<span class="hljs-keyword">extern</span> <span class="hljs-keyword">crate</span> tempdir;
<span class="hljs-meta">#[macro_use]</span>
<span class="hljs-keyword">extern</span> <span class="hljs-keyword">crate</span> serde_json;
<span class="hljs-keyword">use</span> std::path::Path;
<span class="hljs-keyword">use</span> tempdir::TempDir;
<span class="hljs-keyword">use</span> tantivy::Index;
@@ -52,7 +54,7 @@
<div class="pilwrap ">
<a class="pilcrow" href="#section-2">&#182;</a>
</div>
<p>Let’s create a temporary directory for the
<p>Let’s create a temporary directory for the
sake of this example</p>
</div>
@@ -60,7 +62,7 @@ sake of this example</p>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">if</span> <span class="hljs-keyword">let</span> <span class="hljs-literal">Ok</span>(dir) = TempDir::new(<span class="hljs-string">"tantivy_example_dir"</span>) {
run_example(dir.path()).unwrap();
dir.close().unwrap();
}
}
}
@@ -78,7 +80,7 @@ sake of this example</p>
<h1 id="defining-the-schema">Defining the schema</h1>
<p>The Tantivy index requires a very strict schema.
The schema declares which fields are in the index,
and for each field, its type and “the way it should
and for each field, its type and “the way it should
be indexed”.</p>
</div>
@@ -108,15 +110,15 @@ be indexed”.</p>
<a class="pilcrow" href="#section-5">&#182;</a>
</div>
<p>Our first field is title.
We want full-text search for it, and we want to be able
to retrieve the document after the search.</p>
We want full-text search for it, and we also want
to be able to retrieve the document after the search.</p>
<p>TEXT | STORED is some syntactic sugar to describe
that. </p>
that.</p>
<p><code>TEXT</code> means the field should be tokenized and indexed,
along with its term frequency and term positions.</p>
<p><code>STORED</code> means that the field will also be saved
in a compressed, row-oriented key-value store.
This store is useful to reconstruct the
This store is useful to reconstruct the
documents that were selected during the search phase.</p>
</div>
@@ -132,14 +134,17 @@ documents that were selected during the search phase.</p>
<div class="pilwrap ">
<a class="pilcrow" href="#section-6">&#182;</a>
</div>
<p>Our first field is body.
We want full-text search for it, and we want to be able
to retrieve the body after the search.</p>
<p>Our second field is body.
We want full-text search for it, but we do not
need to be able to be able to retrieve it
for our application. </p>
<p>We can make our index lighter and
by omitting <code>STORED</code> flag.</p>
</div>
<div class="content"><div class='highlight'><pre> schema_builder.add_text_field(<span class="hljs-string">"body"</span>, TEXT);
<span class="hljs-keyword">let</span> schema = schema_builder.build();</pre></div></div>
</li>
@@ -158,7 +163,7 @@ with our schema in the directory.</p>
</div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> index = <span class="hljs-built_in">try!</span>(Index::create(index_path, schema.clone()));</pre></div></div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> index = Index::create(index_path, schema.clone())?;</pre></div></div>
</li>
@@ -173,14 +178,12 @@ with our schema in the directory.</p>
There must be only one writer at a time.
This single <code>IndexWriter</code> is already
multithreaded.</p>
<p>Here we use a buffer of 1 GB. Using a bigger
heap for the indexer can increase its throughput.
This buffer will be split between the indexing
threads.</p>
<p>Here we use a buffer of 50MB per thread. Using a bigger
heap for the indexer can increase its throughput.</p>
</div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> <span class="hljs-keyword">mut</span> index_writer = <span class="hljs-built_in">try!</span>(index.writer(<span class="hljs-number">1_000_000_000</span>));</pre></div></div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> <span class="hljs-keyword">mut</span> index_writer = index.writer(<span class="hljs-number">50_000_000</span>)?;</pre></div></div>
</li>
@@ -213,10 +216,14 @@ one by one in a Document object.</p>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> title = schema.get_field(<span class="hljs-string">"title"</span>).unwrap();
<span class="hljs-keyword">let</span> body = schema.get_field(<span class="hljs-string">"body"</span>).unwrap();
<span class="hljs-keyword">let</span> <span class="hljs-keyword">mut</span> old_man_doc = Document::<span class="hljs-keyword">default</span>();
old_man_doc.add_text(title, <span class="hljs-string">"The Old Man and the Sea"</span>);
old_man_doc.add_text(body, <span class="hljs-string">"He was an old man who fished alone in a skiff in the Gulf Stream and he had gone eighty-four days now without taking a fish."</span>);</pre></div></div>
old_man_doc.add_text(
body,
<span class="hljs-string">"He was an old man who fished alone in a skiff in the Gulf Stream and \
he had gone eighty-four days now without taking a fish."</span>,
);</pre></div></div>
</li>
@@ -231,7 +238,7 @@ one by one in a Document object.</p>
</div>
<div class="content"><div class='highlight'><pre> <span class="hljs-built_in">try!</span>(index_writer.add_document(old_man_doc));</pre></div></div>
<div class="content"><div class='highlight'><pre> index_writer.add_document(old_man_doc);</pre></div></div>
</li>
@@ -243,18 +250,27 @@ one by one in a Document object.</p>
<a class="pilcrow" href="#section-12">&#182;</a>
</div>
<h3 id="create-a-document-directly-from-json-">Create a document directly from json.</h3>
<p>Alternatively, we can use our schema to parse
a document object directly from json.</p>
<p>Alternatively, we can use our schema to parse a
document object directly from json.
The document is a string, but we use the <code>json</code> macro
from <code>serde_json</code> for the convenience of multi-line support.</p>
</div>
<div class="content"><div class='highlight'><pre>
<span class="hljs-keyword">let</span> mice_and_men_doc = <span class="hljs-built_in">try!</span>(schema.parse_document(r#<span class="hljs-string">"{
"</span>title<span class="hljs-string">": "</span>Of Mice and Men<span class="hljs-string">",
"</span>body<span class="hljs-string">": "</span>few miles south of Soledad, the Salinas River drops <span class="hljs-keyword">in</span> close to the hillside bank and runs deep and green. The water is warm too, <span class="hljs-keyword">for</span> it has slipped twinkling over the yellow sands <span class="hljs-keyword">in</span> the sunlight before reaching the narrow pool. On one side of the river the golden foothill slopes curve up to the strong and rocky Gabilan Mountains, but on the valley side the water is lined with trees—willows fresh and green with every spring, carrying <span class="hljs-keyword">in</span> their lower leaf junctures the debris of the winter’s flooding; and sycamores with mottled, white,recumbent limbs and branches that arch over the pool<span class="hljs-string">"
}"</span>#));
<span class="hljs-built_in">try!</span>(index_writer.add_document(mice_and_men_doc));</pre></div></div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> json = json!({
<span class="hljs-string">"title"</span>: <span class="hljs-string">"Of Mice and Men"</span>,
<span class="hljs-string">"body"</span>: <span class="hljs-string">"A few miles south of Soledad, the Salinas River drops in close to the hillside \
bank and runs deep and green. The water is warm too, for it has slipped twinkling \
over the yellow sands in the sunlight before reaching the narrow pool. On one \
side of the river the golden foothill slopes curve up to the strong and rocky \
Gabilan Mountains, but on the valley side the water is lined with trees—willows \
fresh and green with every spring, carrying in their lower leaf junctures the \
debris of the winter’s flooding; and sycamores with mottled, white, recumbent \
limbs and branches that arch over the pool"</span>
});
<span class="hljs-keyword">let</span> mice_and_men_doc = schema.parse_document(&amp;json.to_string())?;
index_writer.add_document(mice_and_men_doc);</pre></div></div>
</li>
@@ -271,11 +287,16 @@ The following document has two titles.</p>
</div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> frankenstein_doc = <span class="hljs-built_in">try!</span>(schema.parse_document(r#<span class="hljs-string">"{
"</span>title<span class="hljs-string">": ["</span>Frankenstein<span class="hljs-string">", "</span>The Modern Promotheus<span class="hljs-string">"],
"</span>body<span class="hljs-string">": "</span>You will rejoice to hear that no disaster has accompanied the commencement of an enterprise which you have regarded with such evil forebodings. I arrived here yesterday, and my first task is to assure my dear sister of my welfare and increasing confidence <span class="hljs-keyword">in</span> the success of my undertaking.<span class="hljs-string">"
}"</span>#));
<span class="hljs-built_in">try!</span>(index_writer.add_document(frankenstein_doc));</pre></div></div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> json = json!({
<span class="hljs-string">"title"</span>: [<span class="hljs-string">"Frankenstein"</span>, <span class="hljs-string">"The Modern Prometheus"</span>],
<span class="hljs-string">"body"</span>: <span class="hljs-string">"You will rejoice to hear that no disaster has accompanied the commencement of an \
enterprise which you have regarded with such evil forebodings. I arrived here \
yesterday, and my first task is to assure my dear sister of my welfare and \
increasing confidence in the success of my undertaking."</span>
});
<span class="hljs-keyword">let</span> frankenstein_doc = schema.parse_document(&amp;json.to_string())?;
index_writer.add_document(frankenstein_doc);</pre></div></div>
</li>
@@ -288,7 +309,7 @@ The following document has two titles.</p>
</div>
<p>This is an example, so we will only index 3 documents
here. You can check out tantivy’s tutorial to index
the English wikipedia. Tantivy’s indexing is rather fast.
the English wikipedia. Tantivy’s indexing is rather fast.
Indexing 5 million articles of the English wikipedia takes
around 4 minutes on my computer!</p>
@@ -313,7 +334,7 @@ the existence of new documents.</p>
</div>
<div class="content"><div class='highlight'><pre> <span class="hljs-built_in">try!</span>(index_writer.commit());</pre></div></div>
<div class="content"><div class='highlight'><pre> index_writer.commit()?;</pre></div></div>
</li>
@@ -343,15 +364,13 @@ commit.</p>
<a class="pilcrow" href="#section-17">&#182;</a>
</div>
<h1 id="searching">Searching</h1>
<p>Let’s search our index. We start
by creating a searcher. There can be more
than one searcher at a time.</p>
<p>You should create a searcher
every time you start a “search query”.</p>
<p>Let’s search our index. Start by reloading
searchers in the index. This should be done
after every commit().</p>
</div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> searcher = index.searcher();</pre></div></div>
<div class="content"><div class='highlight'><pre> index.load_searchers()?;</pre></div></div>
</li>
@@ -362,14 +381,13 @@ every time you start a “search query”.</p>
<div class="pilwrap ">
<a class="pilcrow" href="#section-18">&#182;</a>
</div>
<p>The query parser can interpret human queries.
Here, if the user does not specify which
field they want to search, tantivy will search
in both title and body.</p>
<p>Afterwards create one (or more) searchers.</p>
<p>You should create a searcher
every time you start a “search query”.</p>
</div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> query_parser = QueryParser::new(index.schema(), <span class="hljs-built_in">vec!</span>(title, body));</pre></div></div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> searcher = index.searcher();</pre></div></div>
</li>
@@ -380,13 +398,14 @@ in both title and body.</p>
<div class="pilwrap ">
<a class="pilcrow" href="#section-19">&#182;</a>
</div>
<p>QueryParser may fail if the query is not in the right
format. For user facing applications, this can be a problem.
A ticket has been opened regarding this problem.</p>
<p>The query parser can interpret human queries.
Here, if the user does not specify which
field they want to search, tantivy will search
in both title and body.</p>
</div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> query = <span class="hljs-built_in">try!</span>(query_parser.parse_query(<span class="hljs-string">"sea whale"</span>));</pre></div></div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> <span class="hljs-keyword">mut</span> query_parser = QueryParser::for_index(index, <span class="hljs-built_in">vec!</span>[title, body]);</pre></div></div>
</li>
@@ -397,6 +416,23 @@ A ticket has been opened regarding this problem.</p>
<div class="pilwrap ">
<a class="pilcrow" href="#section-20">&#182;</a>
</div>
<p>QueryParser may fail if the query is not in the right
format. For user facing applications, this can be a problem.
A ticket has been opened regarding this problem.</p>
</div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> query = query_parser.parse_query(<span class="hljs-string">"sea whale"</span>)?;</pre></div></div>
</li>
<li id="section-21">
<div class="annotation">
<div class="pilwrap ">
<a class="pilcrow" href="#section-21">&#182;</a>
</div>
<p>A query defines a set of documents, as
well as the way they should be scored.</p>
<p>A query created by the query parser is scored according
@@ -408,36 +444,20 @@ any document matching at least one of our terms.</p>
</li>
<li id="section-21">
<div class="annotation">
<div class="pilwrap ">
<a class="pilcrow" href="#section-21">&#182;</a>
</div>
<h3 id="collectors">Collectors</h3>
<p>We are not interested in all of the documents but
only in the top 10. Keeping track of our top 10 best documents
is the role of the TopCollector.</p>
</div>
<div class="content"><div class='highlight'><pre>
<span class="hljs-keyword">let</span> <span class="hljs-keyword">mut</span> top_collector = TopCollector::with_limit(<span class="hljs-number">10</span>);</pre></div></div>
</li>
<li id="section-22">
<div class="annotation">
<div class="pilwrap ">
<a class="pilcrow" href="#section-22">&#182;</a>
</div>
<p>We can now perform our query.</p>
<h3 id="collectors">Collectors</h3>
<p>We are not interested in all of the documents but
only in the top 10. Keeping track of our top 10 best documents
is the role of the TopCollector.</p>
</div>
<div class="content"><div class='highlight'><pre> <span class="hljs-built_in">try!</span>(searcher.search(&amp;query, &amp;<span class="hljs-keyword">mut</span> top_collector)));</pre></div></div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> <span class="hljs-keyword">mut</span> top_collector = TopCollector::with_limit(<span class="hljs-number">10</span>);</pre></div></div>
</li>
@@ -448,12 +468,11 @@ is the role of the TopCollector.</p>
<div class="pilwrap ">
<a class="pilcrow" href="#section-23">&#182;</a>
</div>
<p>Our top collector now contains the 10
most relevant doc ids…</p>
<p>We can now perform our query.</p>
</div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> doc_addresses = top_collector.docs();</pre></div></div>
<div class="content"><div class='highlight'><pre> searcher.search(&amp;*query, &amp;<span class="hljs-keyword">mut</span> top_collector)?;</pre></div></div>
</li>
@@ -464,7 +483,23 @@ most relevant doc ids…</p>
<div class="pilwrap ">
<a class="pilcrow" href="#section-24">&#182;</a>
</div>
<p>The actual documents still need to be
<p>Our top collector now contains the 10
most relevant doc ids…</p>
</div>
<div class="content"><div class='highlight'><pre> <span class="hljs-keyword">let</span> doc_addresses = top_collector.docs();</pre></div></div>
</li>
<li id="section-25">
<div class="annotation">
<div class="pilwrap ">
<a class="pilcrow" href="#section-25">&#182;</a>
</div>
<p>The actual documents still need to be
retrieved from Tantivy’s store.</p>
<p>Since the body field was not configured as stored,
the document returned will only contain
@@ -472,11 +507,29 @@ a title.</p>
</div>
<div class="content"><div class='highlight'><pre>
<div class="content"><div class='highlight'><pre>
<span class="hljs-keyword">for</span> doc_address <span class="hljs-keyword">in</span> doc_addresses {
<span class="hljs-keyword">let</span> retrieved_doc = <span class="hljs-built_in">try!</span>(searcher.doc(&amp;doc_address));
<span class="hljs-built_in">println!</span>(<span class="hljs-string">"{}"</span>, schema.to_json(&amp;retrieved_doc));
}
<span class="hljs-keyword">let</span> retrieved_doc = searcher.doc(&amp;doc_address)?;
<span class="hljs-built_in">println!</span>(<span class="hljs-string">"{}"</span>, schema.to_json(&amp;retrieved_doc));
}</pre></div></div>
</li>
<li id="section-26">
<div class="annotation">
<div class="pilwrap ">
<a class="pilcrow" href="#section-26">&#182;</a>
</div>
<p>Wait for indexing and merging threads to shut down.
Usually this isn’t needed, but in <code>main</code> we try to
delete the temporary directory and that fails on
Windows if the files are still open.</p>
</div>
<div class="content"><div class='highlight'><pre> index_writer.wait_merging_threads()?;
<span class="hljs-literal">Ok</span>(())
}</pre></div></div>

188
examples/simple_search.rs
View File

@@ -1,7 +1,9 @@
extern crate rustc_serialize;
extern crate tantivy;
extern crate tempdir;
#[macro_use]
extern crate serde_json;
use std::path::Path;
use tempdir::TempDir;
use tantivy::Index;
@@ -10,149 +12,161 @@ use tantivy::collector::TopCollector;
use tantivy::query::QueryParser;
fn main() {
// Let's create a temporary directory for the
// Let's create a temporary directory for the
// sake of this example
if let Ok(dir) = TempDir::new("tantivy_example_dir") {
run_example(dir.path()).unwrap();
dir.close().unwrap();
}
}
}
fn run_example(index_path: &Path) -> tantivy::Result<()> {
// # Defining the schema
//
// The Tantivy index requires a very strict schema.
// The schema declares which fields are in the index,
// and for each field, its type and "the way it should
// and for each field, its type and "the way it should
// be indexed".
// first we need to define a schema ...
let mut schema_builder = SchemaBuilder::default();
// Our first field is title.
// We want full-text search for it, and we want to be able
// to retrieve the document after the search.
// We want full-text search for it, and we also want
// to be able to retrieve the document after the search.
//
// TEXT | STORED is some syntactic sugar to describe
// that.
//
// that.
//
// `TEXT` means the field should be tokenized and indexed,
// along with its term frequency and term positions.
//
// `STORED` means that the field will also be saved
// in a compressed, row-oriented key-value store.
// This store is useful to reconstruct the
// This store is useful to reconstruct the
// documents that were selected during the search phase.
schema_builder.add_text_field("title", TEXT | STORED);
// Our first field is body.
// We want full-text search for it, and we want to be able
// to retrieve the body after the search.
// Our second field is body.
// We want full-text search for it, but we do not
// need to be able to be able to retrieve it
// for our application.
//
// We can make our index lighter and
// by omitting `STORED` flag.
schema_builder.add_text_field("body", TEXT);
let schema = schema_builder.build();
let schema = schema_builder.build();
// # Indexing documents
//
// Let's create a brand new index.
//
//
// This will actually just save a meta.json
// with our schema in the directory.
let index = try!(Index::create(index_path, schema.clone()));
let index = Index::create(index_path, schema.clone())?;
// To insert document we need an index writer.
// There must be only one writer at a time.
// This single `IndexWriter` is already
// multithreaded.
//
// Here we use a buffer of 1 GB. Using a bigger
// Here we use a buffer of 50MB per thread. Using a bigger
// heap for the indexer can increase its throughput.
// This buffer will be split between the indexing
// threads.
let mut index_writer = try!(index.writer(1_000_000_000));
let mut index_writer = index.writer(50_000_000)?;
// Let's index our documents!
// We first need a handle on the title and the body field.
// ### Create a document "manually".
//
// We can create a document manually, by setting the fields
// one by one in a Document object.
let title = schema.get_field("title").unwrap();
let body = schema.get_field("body").unwrap();
let mut old_man_doc = Document::default();
old_man_doc.add_text(title, "The Old Man and the Sea");
old_man_doc.add_text(body, "He was an old man who fished alone in a skiff in the Gulf Stream and he had gone eighty-four days now without taking a fish.");
old_man_doc.add_text(
body,
"He was an old man who fished alone in a skiff in the Gulf Stream and \
he had gone eighty-four days now without taking a fish.",
);
// ... and add it to the `IndexWriter`.
try!(index_writer.add_document(old_man_doc));
index_writer.add_document(old_man_doc);
// ### Create a document directly from json.
//
// Alternatively, we can use our schema to parse
// a document object directly from json.
let mice_and_men_doc = try!(schema.parse_document(r#"{
// Alternatively, we can use our schema to parse a
// document object directly from json.
// The document is a string, but we use the `json` macro
// from `serde_json` for the convenience of multi-line support.
let json = json!({
"title": "Of Mice and Men",
"body": "few miles south of Soledad, the Salinas River drops in close to the hillside bank and runs deep and green. The water is warm too, for it has slipped twinkling over the yellow sands in the sunlight before reaching the narrow pool. On one side of the river the golden foothill slopes curve up to the strong and rocky Gabilan Mountains, but on the valley side the water is lined with trees—willows fresh and green with every spring, carrying in their lower leaf junctures the debris of the winter’s flooding; and sycamores with mottled, white,recumbent limbs and branches that arch over the pool"
}"#));
try!(index_writer.add_document(mice_and_men_doc));
"body": "A few miles south of Soledad, the Salinas River drops in close to the hillside \
bank and runs deep and green. The water is warm too, for it has slipped twinkling \
over the yellow sands in the sunlight before reaching the narrow pool. On one \
side of the river the golden foothill slopes curve up to the strong and rocky \
Gabilan Mountains, but on the valley side the water is lined with trees—willows \
fresh and green with every spring, carrying in their lower leaf junctures the \
debris of the winter’s flooding; and sycamores with mottled, white, recumbent \
limbs and branches that arch over the pool"
});
let mice_and_men_doc = schema.parse_document(&json.to_string())?;
index_writer.add_document(mice_and_men_doc);
// Multi-valued field are allowed, they are
// expressed in JSON by an array.
// The following document has two titles.
let frankenstein_doc = try!(schema.parse_document(r#"{
"title": ["Frankenstein", "The Modern Promotheus"],
"body": "You will rejoice to hear that no disaster has accompanied the commencement of an enterprise which you have regarded with such evil forebodings. I arrived here yesterday, and my first task is to assure my dear sister of my welfare and increasing confidence in the success of my undertaking."
}"#));
try!(index_writer.add_document(frankenstein_doc));
let json = json!({
"title": ["Frankenstein", "The Modern Prometheus"],
"body": "You will rejoice to hear that no disaster has accompanied the commencement of an \
enterprise which you have regarded with such evil forebodings. I arrived here \
yesterday, and my first task is to assure my dear sister of my welfare and \
increasing confidence in the success of my undertaking."
});
let frankenstein_doc = schema.parse_document(&json.to_string())?;
index_writer.add_document(frankenstein_doc);
// This is an example, so we will only index 3 documents
// here. You can check out tantivy's tutorial to index
// the English wikipedia. Tantivy's indexing is rather fast.
// the English wikipedia. Tantivy's indexing is rather fast.
// Indexing 5 million articles of the English wikipedia takes
// around 4 minutes on my computer!
// ### Committing
//
//
// At this point our documents are not searchable.
//
//
//
// We need to call .commit() explicitly to force the
// index_writer to finish processing the documents in the queue,
// flush the current index to the disk, and advertise
// the existence of new documents.
//
// This call is blocking.
try!(index_writer.commit());
index_writer.commit()?;
// If `.commit()` returns correctly, then all of the
// documents that have been added are guaranteed to be
// persistently indexed.
//
//
// In the scenario of a crash or a power failure,
// tantivy behaves as if has rolled back to its last
// commit.
// # Searching
//
// Let's search our index. We start
// by creating a searcher. There can be more
// than one searcher at a time.
//
// Let's search our index. Start by reloading
// searchers in the index. This should be done
// after every commit().
index.load_searchers()?;
// Afterwards create one (or more) searchers.
//
// You should create a searcher
// every time you start a "search query".
let searcher = index.searcher();
@@ -161,47 +175,51 @@ fn run_example(index_path: &Path) -> tantivy::Result<()> {
// Here, if the user does not specify which
// field they want to search, tantivy will search
// in both title and body.
let query_parser = QueryParser::new(index.schema(), vec!(title, body));
let query_parser = QueryParser::for_index(&index, vec![title, body]);
// QueryParser may fail if the query is not in the right
// format. For user facing applications, this can be a problem.
// A ticket has been opened regarding this problem.
let query = try!(query_parser.parse_query("sea whale"));
let query = query_parser.parse_query("sea whale")?;
// A query defines a set of documents, as
// well as the way they should be scored.
//
//
// A query created by the query parser is scored according
// to a metric called Tf-Idf, and will consider
// any document matching at least one of our terms.
// ### Collectors
// ### Collectors
//
// We are not interested in all of the documents but
// We are not interested in all of the documents but
// only in the top 10. Keeping track of our top 10 best documents
// is the role of the TopCollector.
let mut top_collector = TopCollector::with_limit(10);
// We can now perform our query.
try!(searcher.search(&*query, &mut top_collector));
// Our top collector now contains the 10
// We can now perform our query.
searcher.search(&*query, &mut top_collector)?;
// Our top collector now contains the 10
// most relevant doc ids...
let doc_addresses = top_collector.docs();
// The actual documents still need to be
// The actual documents still need to be
// retrieved from Tantivy's store.
//
//
// Since the body field was not configured as stored,
// the document returned will only contain
// a title.
for doc_address in doc_addresses {
let retrieved_doc = try!(searcher.doc(&doc_address));
println!("{}", schema.to_json(&retrieved_doc));
let retrieved_doc = searcher.doc(&doc_address)?;
println!("{}", schema.to_json(&retrieved_doc));
}
// Wait for indexing and merging threads to shut down.
// Usually this isn't needed, but in `main` we try to
// delete the temporary directory and that fails on
// Windows if the files are still open.
index_writer.wait_merging_threads()?;
Ok(())
}

86
src/analyzer/mod.rs
View File

@@ -1,86 +0,0 @@
extern crate regex;
use std::str::Chars;
use std::ascii::AsciiExt;
pub struct TokenIter<'a> {
chars: Chars<'a>,
term_buffer: String,
}
fn append_char_lowercase(c: char, term_buffer: &mut String) {
term_buffer.push(c.to_ascii_lowercase());
}
pub trait StreamingIterator<'a, T> {
fn next(&'a mut self) -> Option<T>;
}
impl<'a, 'b> TokenIter<'b> {
fn consume_token(&'a mut self) -> Option<&'a str> {
for c in &mut self.chars {
if c.is_alphanumeric() {
append_char_lowercase(c, &mut self.term_buffer);
}
else {
break;
}
}
Some(&self.term_buffer)
}
}
impl<'a, 'b> StreamingIterator<'a, &'a str> for TokenIter<'b> {
#[inline]
fn next(&'a mut self,) -> Option<&'a str> {
self.term_buffer.clear();
// skipping non-letter characters.
loop {
match self.chars.next() {
Some(c) => {
if c.is_alphanumeric() {
append_char_lowercase(c, &mut self.term_buffer);
return self.consume_token();
}
}
None => { return None; }
}
}
}
}
pub struct SimpleTokenizer;
impl SimpleTokenizer {
pub fn tokenize<'a>(&self, text: &'a str) -> TokenIter<'a> {
TokenIter {
term_buffer: String::new(),
chars: text.chars(),
}
}
}
#[test]
fn test_tokenizer() {
let simple_tokenizer = SimpleTokenizer;
let mut term_reader = simple_tokenizer.tokenize("hello, happy tax payer!");
assert_eq!(term_reader.next().unwrap(), "hello");
assert_eq!(term_reader.next().unwrap(), "happy");
assert_eq!(term_reader.next().unwrap(), "tax");
assert_eq!(term_reader.next().unwrap(), "payer");
assert_eq!(term_reader.next(), None);
}
#[test]
fn test_tokenizer_empty() {
let simple_tokenizer = SimpleTokenizer;
let mut term_reader = simple_tokenizer.tokenize("");
assert_eq!(term_reader.next(), None);
}

26
src/collector/chained_collector.rs
View File

@@ -1,18 +1,17 @@
use Result;
use collector::Collector;
use SegmentLocalId;
use SegmentReader;
use std::io;
use DocId;
use Score;
/// Collector that does nothing.
/// This is used in the chain Collector and will hopefully
/// be optimized away by the compiler.
/// This is used in the chain Collector and will hopefully
/// be optimized away by the compiler.
pub struct DoNothingCollector;
impl Collector for DoNothingCollector {
#[inline]
fn set_segment(&mut self, _: SegmentLocalId, _: &SegmentReader) -> io::Result<()> {
fn set_segment(&mut self, _: SegmentLocalId, _: &SegmentReader) -> Result<()> {
Ok(())
}
#[inline]
@@ -24,10 +23,10 @@ impl Collector for DoNothingCollector {
/// are known at compile time.
pub struct ChainedCollector<Left: Collector, Right: Collector> {
left: Left,
right: Right
right: Right,
}
impl<Left: Collector, Right: Collector> ChainedCollector<Left, Right> {
impl<Left: Collector, Right: Collector> ChainedCollector<Left, Right> {
/// Adds a collector
pub fn push<C: Collector>(self, new_collector: &mut C) -> ChainedCollector<Self, &mut C> {
ChainedCollector {
@@ -38,7 +37,11 @@ impl<Left: Collector, Right: Collector> ChainedCollector<Left, Right> {
}
impl<Left: Collector, Right: Collector> Collector for ChainedCollector<Left, Right> {
fn set_segment(&mut self, segment_local_id: SegmentLocalId, segment: &SegmentReader) -> io::Result<()> {
fn set_segment(
&mut self,
segment_local_id: SegmentLocalId,
segment: &SegmentReader,
) -> Result<()> {
try!(self.left.set_segment(segment_local_id, segment));
try!(self.right.set_segment(segment_local_id, segment));
Ok(())
@@ -58,7 +61,6 @@ pub fn chain() -> ChainedCollector<DoNothingCollector, DoNothingCollector> {
}
}
#[cfg(test)]
mod tests {
@@ -70,9 +72,7 @@ mod tests {
let mut top_collector = TopCollector::with_limit(2);
let mut count_collector = CountCollector::default();
{
let mut collectors = chain()
.push(&mut top_collector)
.push(&mut count_collector);
let mut collectors = chain().push(&mut top_collector).push(&mut count_collector);
collectors.collect(1, 0.2);
collectors.collect(2, 0.1);
collectors.collect(3, 0.5);
@@ -80,4 +80,4 @@ mod tests {
assert_eq!(count_collector.count(), 3);
assert!(top_collector.at_capacity());
}
}
}

12
src/collector/count_collector.rs
View File

@@ -1,12 +1,12 @@
use std::io;
use super::Collector;
use DocId;
use Score;
use Result;
use SegmentReader;
use SegmentLocalId;
/// `CountCollector` collector only counts how many
/// documents match the query.
/// documents match the query.
pub struct CountCollector {
count: usize,
}
@@ -14,21 +14,19 @@ pub struct CountCollector {
impl CountCollector {
/// Returns the count of documents that were
/// collected.
pub fn count(&self,) -> usize {
pub fn count(&self) -> usize {
self.count
}
}
impl Default for CountCollector {
fn default() -> CountCollector {
CountCollector {count: 0,
}
CountCollector { count: 0 }
}
}
impl Collector for CountCollector {
fn set_segment(&mut self, _: SegmentLocalId, _: &SegmentReader) -> io::Result<()> {
fn set_segment(&mut self, _: SegmentLocalId, _: &SegmentReader) -> Result<()> {
Ok(())
}

113
src/collector/facet_collector.rs Normal file
View File

@@ -0,0 +1,113 @@
use std::cmp::Eq;
use std::collections::HashMap;
use std::hash::Hash;
use collector::Collector;
use fastfield::FastFieldReader;
use schema::Field;
use DocId;
use Result;
use Score;
use SegmentReader;
use SegmentLocalId;
/// Facet collector for i64/u64 fast field
pub struct FacetCollector<T>
where
T: FastFieldReader,
T::ValueType: Eq + Hash,
{
counters: HashMap<T::ValueType, u64>,
field: Field,
ff_reader: Option<T>,
}
impl<T> FacetCollector<T>
where
T: FastFieldReader,
T::ValueType: Eq + Hash,
{
/// Creates a new facet collector for aggregating a given field.
pub fn new(field: Field) -> FacetCollector<T> {
FacetCollector {
counters: HashMap::new(),
field: field,
ff_reader: None,
}
}
}
impl<T> Collector for FacetCollector<T>
where
T: FastFieldReader,
T::ValueType: Eq + Hash,
{
fn set_segment(&mut self, _: SegmentLocalId, reader: &SegmentReader) -> Result<()> {
self.ff_reader = Some(reader.get_fast_field_reader(self.field)?);
Ok(())
}
fn collect(&mut self, doc: DocId, _: Score) {
let val = self.ff_reader
.as_ref()
.expect("collect() was called before set_segment. This should never happen.")
.get(doc);
*(self.counters.entry(val).or_insert(0)) += 1;
}
}
#[cfg(test)]
mod tests {
use collector::{chain, FacetCollector};
use query::QueryParser;
use fastfield::{I64FastFieldReader, U64FastFieldReader};
use schema::{self, FAST, STRING};
use Index;
#[test]
// create 10 documents, set num field value to 0 or 1 for even/odd ones
// make sure we have facet counters correctly filled
fn test_facet_collector_results() {
let mut schema_builder = schema::SchemaBuilder::new();
let num_field_i64 = schema_builder.add_i64_field("num_i64", FAST);
let num_field_u64 = schema_builder.add_u64_field("num_u64", FAST);
let text_field = schema_builder.add_text_field("text", STRING);
let schema = schema_builder.build();
let index = Index::create_in_ram(schema.clone());
{
let mut index_writer = index.writer_with_num_threads(1, 40_000_000).unwrap();
{
for i in 0u64..10u64 {
index_writer.add_document(doc!(
num_field_i64 => ((i as i64) % 3i64) as i64,
num_field_u64 => (i % 2u64) as u64,
text_field => "text"
));
}
}
assert_eq!(index_writer.commit().unwrap(), 10u64);
}
index.load_searchers().unwrap();
let searcher = index.searcher();
let mut ffvf_i64: FacetCollector<I64FastFieldReader> = FacetCollector::new(num_field_i64);
let mut ffvf_u64: FacetCollector<U64FastFieldReader> = FacetCollector::new(num_field_u64);
{
// perform the query
let mut facet_collectors = chain().push(&mut ffvf_i64).push(&mut ffvf_u64);
let query_parser = QueryParser::for_index(&index, vec![text_field]);
let query = query_parser.parse_query("text:text").unwrap();
query.search(&searcher, &mut facet_collectors).unwrap();
}
assert_eq!(ffvf_u64.counters[&0], 5);
assert_eq!(ffvf_u64.counters[&1], 5);
assert_eq!(ffvf_i64.counters[&0], 4);
assert_eq!(ffvf_i64.counters[&1], 3);
}
}

62
src/collector/mod.rs
View File

@@ -1,8 +1,12 @@
/*!
Defines how the documents matching a search query should be processed.
*/
use SegmentReader;
use SegmentLocalId;
use DocId;
use Score;
use std::io;
use Result;
mod count_collector;
pub use self::count_collector::CountCollector;
@@ -13,14 +17,17 @@ pub use self::multi_collector::MultiCollector;
mod top_collector;
pub use self::top_collector::TopCollector;
mod facet_collector;
pub use self::facet_collector::FacetCollector;
mod chained_collector;
pub use self::chained_collector::chain;
/// Collectors are in charge of collecting and retaining relevant
/// Collectors are in charge of collecting and retaining relevant
/// information from the document found and scored by the query.
///
///
/// For instance,
/// For instance,
///
/// - keeping track of the top 10 best documents
/// - computing a breakdown over a fast field
@@ -29,7 +36,7 @@ pub use self::chained_collector::chain;
/// Queries are in charge of pushing the `DocSet` to the collector.
///
/// As they work on multiple segments, they first inform
/// the collector of a change in a segment and then
/// the collector of a change in a segment and then
/// call the `collect` method to push the document to the collector.
///
/// Temporally, our collector will receive calls
@@ -46,16 +53,23 @@ pub use self::chained_collector::chain;
///
/// Segments are not guaranteed to be visited in any specific order.
pub trait Collector {
/// `set_segment` is called before beginning to enumerate
/// `set_segment` is called before beginning to enumerate
/// on this segment.
fn set_segment(&mut self, segment_local_id: SegmentLocalId, segment: &SegmentReader) -> io::Result<()>;
fn set_segment(
&mut self,
segment_local_id: SegmentLocalId,
segment: &SegmentReader,
) -> Result<()>;
/// The query pushes the scored document to the collector via this method.
fn collect(&mut self, doc: DocId, score: Score);
}
impl<'a, C: Collector> Collector for &'a mut C {
fn set_segment(&mut self, segment_local_id: SegmentLocalId, segment: &SegmentReader) -> io::Result<()> {
fn set_segment(
&mut self,
segment_local_id: SegmentLocalId,
segment: &SegmentReader,
) -> Result<()> {
(*self).set_segment(segment_local_id, segment)
}
/// The query pushes the scored document to the collector via this method.
@@ -64,7 +78,6 @@ impl<'a, C: Collector> Collector for &'a mut C {
}
}
#[cfg(test)]
pub mod tests {
@@ -73,11 +86,11 @@ pub mod tests {
use DocId;
use Score;
use core::SegmentReader;
use std::io;
use SegmentLocalId;
use fastfield::U32FastFieldReader;
use fastfield::U64FastFieldReader;
use fastfield::FastFieldReader;
use schema::Field;
/// Stores all of the doc ids.
/// This collector is only used for tests.
/// It is unusable in practise, as it does not store
@@ -90,7 +103,7 @@ pub mod tests {
impl TestCollector {
/// Return the exhalist of documents.
pub fn docs(self,) -> Vec<DocId> {
pub fn docs(self) -> Vec<DocId> {
self.docs
}
}
@@ -106,8 +119,7 @@ pub mod tests {
}
impl Collector for TestCollector {
fn set_segment(&mut self, _: SegmentLocalId, reader: &SegmentReader) -> io::Result<()> {
fn set_segment(&mut self, _: SegmentLocalId, reader: &SegmentReader) -> Result<()> {
self.offset += self.segment_max_doc;
self.segment_max_doc = reader.max_doc();
Ok(())
@@ -117,18 +129,15 @@ pub mod tests {
self.docs.push(doc + self.offset);
}
}
/// Collects in order all of the fast fields for all of the
/// doc in the `DocSet`
///
/// This collector is mainly useful for tests.
pub struct FastFieldTestCollector {
vals: Vec<u32>,
vals: Vec<u64>,
field: Field,
ff_reader: Option<U32FastFieldReader>,
ff_reader: Option<U64FastFieldReader>,
}
impl FastFieldTestCollector {
@@ -140,14 +149,14 @@ pub mod tests {
}
}
pub fn vals(&self,) -> &Vec<u32> {
&self.vals
pub fn vals(self) -> Vec<u64> {
self.vals
}
}
impl Collector for FastFieldTestCollector {
fn set_segment(&mut self, _: SegmentLocalId, reader: &SegmentReader) -> io::Result<()> {
self.ff_reader = Some(try!(reader.get_fast_field_reader(self.field)));
fn set_segment(&mut self, _: SegmentLocalId, reader: &SegmentReader) -> Result<()> {
self.ff_reader = Some(reader.get_fast_field_reader(self.field)?);
Ok(())
}
@@ -157,7 +166,6 @@ pub mod tests {
}
}
#[bench]
fn build_collector(b: &mut Bencher) {
b.iter(|| {

17
src/collector/multi_collector.rs
View File

@@ -1,13 +1,12 @@
use std::io;
use super::Collector;
use DocId;
use Score;
use Result;
use SegmentReader;
use SegmentLocalId;
/// Multicollector makes it possible to collect on more than one collector.
/// It should only be used for use cases where the Collector types is unknown
/// It should only be used for use cases where the Collector types is unknown
/// at compile time.
/// If the type of the collectors is known, you should prefer to use `ChainedCollector`.
pub struct MultiCollector<'a> {
@@ -23,9 +22,12 @@ impl<'a> MultiCollector<'a> {
}
}
impl<'a> Collector for MultiCollector<'a> {
fn set_segment(&mut self, segment_local_id: SegmentLocalId, segment: &SegmentReader) -> io::Result<()> {
fn set_segment(
&mut self,
segment_local_id: SegmentLocalId,
segment: &SegmentReader,
) -> Result<()> {
for collector in &mut self.collectors {
try!(collector.set_segment(segment_local_id, segment));
}
@@ -39,8 +41,6 @@ impl<'a> Collector for MultiCollector<'a> {
}
}
#[cfg(test)]
mod tests {
@@ -52,7 +52,8 @@ mod tests {
let mut top_collector = TopCollector::with_limit(2);
let mut count_collector = CountCollector::default();
{
let mut collectors = MultiCollector::from(vec!(&mut top_collector, &mut count_collector));
let mut collectors =
MultiCollector::from(vec![&mut top_collector, &mut count_collector]);
collectors.collect(1, 0.2);
collectors.collect(2, 0.1);
collectors.collect(3, 0.5);

68
src/collector/top_collector.rs
View File

@@ -1,8 +1,8 @@
use std::io;
use super::Collector;
use SegmentReader;
use SegmentLocalId;
use DocAddress;
use Result;
use std::collections::BinaryHeap;
use std::cmp::Ordering;
use DocId;
@@ -12,8 +12,7 @@ use Score;
#[derive(Clone, Copy)]
struct GlobalScoredDoc {
score: Score,
doc_address: DocAddress
doc_address: DocAddress,
}
impl PartialOrd for GlobalScoredDoc {
@@ -25,10 +24,10 @@ impl PartialOrd for GlobalScoredDoc {
impl Ord for GlobalScoredDoc {
#[inline]
fn cmp(&self, other: &GlobalScoredDoc) -> Ordering {
other.score.partial_cmp(&self.score)
.unwrap_or(
other.doc_address.cmp(&self.doc_address)
)
other
.score
.partial_cmp(&self.score)
.unwrap_or_else(|| other.doc_address.cmp(&self.doc_address))
}
}
@@ -40,7 +39,6 @@ impl PartialEq for GlobalScoredDoc {
impl Eq for GlobalScoredDoc {}
/// The Top Collector keeps track of the K documents
/// with the best scores.
///
@@ -53,7 +51,6 @@ pub struct TopCollector {
}
impl TopCollector {
/// Creates a top collector, with a number of documents equal to "limit".
///
/// # Panics
@@ -68,9 +65,9 @@ impl TopCollector {
segment_id: 0,
}
}
/// Returns K best documents sorted in decreasing order.
///
///
/// Calling this method triggers the sort.
/// The result of the sort is not cached.
pub fn docs(&self) -> Vec<DocAddress> {
@@ -81,31 +78,28 @@ impl TopCollector {
}
/// Returns K best ScoredDocument sorted in decreasing order.
///
///
/// Calling this method triggers the sort.
/// The result of the sort is not cached.
pub fn score_docs(&self) -> Vec<(Score, DocAddress)> {
let mut scored_docs: Vec<GlobalScoredDoc> = self.heap
.iter()
.cloned()
.collect();
let mut scored_docs: Vec<GlobalScoredDoc> = self.heap.iter().cloned().collect();
scored_docs.sort();
scored_docs.into_iter()
.map(|GlobalScoredDoc {score, doc_address}| (score, doc_address))
scored_docs
.into_iter()
.map(|GlobalScoredDoc { score, doc_address }| (score, doc_address))
.collect()
}
/// Return true iff at least K documents have gone through
/// the collector.
#[inline]
pub fn at_capacity(&self, ) -> bool {
pub fn at_capacity(&self) -> bool {
self.heap.len() >= self.limit
}
}
impl Collector for TopCollector {
fn set_segment(&mut self, segment_id: SegmentLocalId, _: &SegmentReader) -> io::Result<()> {
fn set_segment(&mut self, segment_id: SegmentLocalId, _: &SegmentReader) -> Result<()> {
self.segment_id = segment_id;
Ok(())
}
@@ -113,25 +107,26 @@ impl Collector for TopCollector {
fn collect(&mut self, doc: DocId, score: Score) {
if self.at_capacity() {
// It's ok to unwrap as long as a limit of 0 is forbidden.
let limit_doc: GlobalScoredDoc = *self.heap.peek().expect("Top collector with size 0 is forbidden");
let limit_doc: GlobalScoredDoc = *self.heap
.peek()
.expect("Top collector with size 0 is forbidden");
if limit_doc.score < score {
let mut mut_head = self.heap.peek_mut().expect("Top collector with size 0 is forbidden");
let mut mut_head = self.heap
.peek_mut()
.expect("Top collector with size 0 is forbidden");
mut_head.score = score;
mut_head.doc_address = DocAddress(self.segment_id, doc);
mut_head.doc_address = DocAddress(self.segment_id, doc);
}
}
else {
} else {
let wrapped_doc = GlobalScoredDoc {
score: score,
doc_address: DocAddress(self.segment_id, doc)
doc_address: DocAddress(self.segment_id, doc),
};
self.heap.push(wrapped_doc);
}
}
}
#[cfg(test)]
mod tests {
@@ -147,13 +142,12 @@ mod tests {
top_collector.collect(3, 0.2);
top_collector.collect(5, 0.3);
assert!(!top_collector.at_capacity());
let score_docs: Vec<(Score, DocId)> = top_collector.score_docs()
let score_docs: Vec<(Score, DocId)> = top_collector
.score_docs()
.into_iter()
.map(|(score, doc_address)| (score, doc_address.doc()))
.collect();
assert_eq!(score_docs, vec!(
(0.8, 1), (0.3, 5), (0.2, 3),
));
assert_eq!(score_docs, vec![(0.8, 1), (0.3, 5), (0.2, 3)]);
}
#[test]
@@ -171,9 +165,7 @@ mod tests {
.into_iter()
.map(|(score, doc_address)| (score, doc_address.doc()))
.collect();
assert_eq!(score_docs, vec!(
(0.9, 7), (0.8, 1), (0.3, 5), (0.2, 3)
));
assert_eq!(score_docs, vec![(0.9, 7), (0.8, 1), (0.3, 5), (0.2, 3)]);
}
{
let docs: Vec<DocId> = top_collector
@@ -181,10 +173,8 @@ mod tests {
.into_iter()
.map(|doc_address| doc_address.doc())
.collect();
assert_eq!(docs, vec!(7, 1, 5, 3));
assert_eq!(docs, vec![7, 1, 5, 3]);
}
}
#[test]

224
src/common/bitpacker.rs
View File

@@ -2,118 +2,181 @@ use std::io::Write;
use std::io;
use common::serialize::BinarySerializable;
use std::mem;
use std::ops::Deref;
pub fn compute_num_bits(amplitude: u32) -> u8 {
(32u32 - amplitude.leading_zeros()) as u8
/// Computes the number of bits that will be used for bitpacking.
///
/// In general the target is the minimum number of bits
/// required to express the amplitude given in argument.
///
/// e.g. If the amplitude is 10, we can store all ints on simply 4bits.
///
/// The logic is slightly more convoluted here as for optimization
/// reasons, we want to ensure that a value spawns over at most 8 bytes
/// of aligns bytes.
///
/// Spanning over 9 bytes is possible for instance, if we do
/// bitpacking with an amplitude of 63 bits.
/// In this case, the second int will start on bit
/// 63 (which belongs to byte 7) and ends at byte 15;
/// Hence 9 bytes (from byte 7 to byte 15 included).
///
/// To avoid this, we force the number of bits to 64bits
/// when the result is greater than `64-8 = 56 bits`.
///
/// Note that this only affects rare use cases spawning over
/// a very large range of values. Even in this case, it results
/// in an extra cost of at most 12% compared to the optimal
/// number of bits.
pub fn compute_num_bits(amplitude: u64) -> u8 {
let amplitude = (64u32 - amplitude.leading_zeros()) as u8;
if amplitude <= 64 - 8 {
amplitude
} else {
64
}
}
pub struct BitPacker {
mini_buffer: u64,
mini_buffer_written: usize,
num_bits: usize,
written_size: usize,
}
impl BitPacker {
impl BitPacker {
pub fn new(num_bits: usize) -> BitPacker {
BitPacker {
mini_buffer: 0u64,
mini_buffer_written: 0,
num_bits: num_bits,
written_size: 0,
num_bits,
}
}
pub fn write<TWrite: Write>(&mut self, val: u32, output: &mut TWrite) -> io::Result<()> {
pub fn write<TWrite: Write>(&mut self, val: u64, output: &mut TWrite) -> io::Result<()> {
let val_u64 = val as u64;
if self.mini_buffer_written + self.num_bits > 64 {
self.mini_buffer |= val_u64.wrapping_shl(self.mini_buffer_written as u32);
self.written_size += self.mini_buffer.serialize(output)?;
self.mini_buffer.serialize(output)?;
self.mini_buffer = val_u64.wrapping_shr((64 - self.mini_buffer_written) as u32);
self.mini_buffer_written = self.mini_buffer_written + (self.num_bits as usize) - 64;
}
else {
} else {
self.mini_buffer |= val_u64 << self.mini_buffer_written;
self.mini_buffer_written += self.num_bits;
if self.mini_buffer_written == 64 {
self.written_size += self.mini_buffer.serialize(output)?;
self.mini_buffer.serialize(output)?;
self.mini_buffer_written = 0;
self.mini_buffer = 0u64;
}
}
}
Ok(())
}
fn flush<TWrite: Write>(&mut self, output: &mut TWrite) -> io::Result<()>{
pub(crate) fn flush<TWrite: Write>(&mut self, output: &mut TWrite) -> io::Result<()> {
if self.mini_buffer_written > 0 {
let num_bytes = (self.mini_buffer_written + 7) / 8;
let arr: [u8; 8] = unsafe { mem::transmute::<u64, [u8; 8]>(self.mini_buffer) };
let arr: [u8; 8] = unsafe { mem::transmute::<u64, [u8; 8]>(self.mini_buffer) };
output.write_all(&arr[..num_bytes])?;
self.written_size += num_bytes;
self.mini_buffer_written = 0;
}
Ok(())
}
pub fn close<TWrite: Write>(&mut self, output: &mut TWrite) -> io::Result<usize> {
pub fn close<TWrite: Write>(&mut self, output: &mut TWrite) -> io::Result<()> {
self.flush(output)?;
Ok(self.written_size)
// Padding the write file to simplify reads.
output.write_all(&[0u8; 7])?;
Ok(())
}
}
pub struct BitUnpacker {
pub struct BitUnpacker<Data>
where
Data: Deref<Target = [u8]>,
{
num_bits: usize,
mask: u32,
data_ptr: *const u8,
data_len: usize,
mask: u64,
data: Data,
}
impl BitUnpacker {
pub fn new(data: &[u8], num_bits: usize) -> BitUnpacker {
impl<Data> BitUnpacker<Data>
where
Data: Deref<Target = [u8]>,
{
pub fn new(data: Data, num_bits: usize) -> BitUnpacker<Data> {
let mask: u64 = if num_bits == 64 {
!0u64
} else {
(1u64 << num_bits) - 1u64
};
BitUnpacker {
num_bits: num_bits,
mask: (1u32 << num_bits) - 1u32,
data_ptr: data.as_ptr(),
data_len: data.len()
num_bits,
mask,
data,
}
}
pub fn get(&self, idx: usize) -> u32 {
pub fn get(&self, idx: usize) -> u64 {
if self.num_bits == 0 {
return 0;
}
let addr = (idx * self.num_bits) / 8;
let bit_shift = idx * self.num_bits - addr * 8;
let val_unshifted_unmasked: u64;
if addr + 8 <= self.data_len {
val_unshifted_unmasked = unsafe { * (self.data_ptr.offset(addr as isize) as *const u64) };
}
else {
let mut arr = [0u8; 8];
if addr < self.data_len {
for i in 0..self.data_len - addr {
arr[i] = unsafe { *self.data_ptr.offset( (addr + i) as isize) };
let data: &[u8] = &*self.data;
let num_bits = self.num_bits;
let mask = self.mask;
let addr_in_bits = idx * num_bits;
let addr = addr_in_bits >> 3;
let bit_shift = addr_in_bits & 7;
if cfg!(feature = "simdcompression") {
// for simdcompression,
// the bitpacker is only used for fastfields,
// and we expect them to be always padded.
debug_assert!(
addr + 8 <= data.len(),
"The fast field field should have been padded with 7 bytes."
);
let val_unshifted_unmasked: u64 = unsafe { *(data[addr..].as_ptr() as *const u64) };
let val_shifted = (val_unshifted_unmasked >> bit_shift) as u64;
(val_shifted & mask)
} else {
let val_unshifted_unmasked: u64 = if addr + 8 <= data.len() {
unsafe { *(data[addr..].as_ptr() as *const u64) }
} else {
let mut buffer = [0u8; 8];
for i in addr..data.len() {
buffer[i - addr] += data[i];
}
}
val_unshifted_unmasked = unsafe { mem::transmute::<[u8; 8], u64>(arr) };
unsafe { *(buffer[..].as_ptr() as *const u64) }
};
let val_shifted = (val_unshifted_unmasked >> bit_shift) as u64;
(val_shifted & mask)
}
}
pub fn get_range(&self, start: u32, output: &mut [u64]) {
if self.num_bits == 0 {
for val in output.iter_mut() {
*val = 0;
}
} else {
let data: &[u8] = &*self.data;
let num_bits = self.num_bits;
let mask = self.mask;
let mut addr_in_bits = (start as usize) * num_bits;
for output_val in output.iter_mut() {
let addr = addr_in_bits >> 3;
let bit_shift = addr_in_bits & 7;
let val_unshifted_unmasked: u64 = unsafe { *(data[addr..].as_ptr() as *const u64) };
let val_shifted = (val_unshifted_unmasked >> bit_shift) as u64;
*output_val = val_shifted & mask;
addr_in_bits += num_bits;
}
}
let val_shifted = (val_unshifted_unmasked >> bit_shift) as u32;
(val_shifted & self.mask)
}
}
#[cfg(test)]
mod test {
use super::{BitPacker, BitUnpacker, compute_num_bits};
use super::{compute_num_bits, BitPacker, BitUnpacker};
#[test]
fn test_compute_num_bits() {
assert_eq!(compute_num_bits(1), 1u8);
@@ -123,32 +186,32 @@ mod test {
assert_eq!(compute_num_bits(4), 3u8);
assert_eq!(compute_num_bits(255), 8u8);
assert_eq!(compute_num_bits(256), 9u8);
assert_eq!(compute_num_bits(5_000_000_000), 33u8);
}
fn test_bitpacker_util(len: usize, num_bits: usize) {
fn create_fastfield_bitpacker(len: usize, num_bits: usize) -> (BitUnpacker<Vec<u8>>, Vec<u64>) {
let mut data = Vec::new();
let mut bitpacker = BitPacker::new(num_bits);
let max_val: u32 = (1 << num_bits) - 1;
let vals: Vec<u32> = (0u32..len as u32).map(|i| {
if max_val == 0 {
0
}
else {
i % max_val
}
}).collect();
let max_val: u64 = (1 << num_bits) - 1;
let vals: Vec<u64> = (0u64..len as u64)
.map(|i| if max_val == 0 { 0 } else { i % max_val })
.collect();
for &val in &vals {
bitpacker.write(val, &mut data).unwrap();
}
let num_bytes = bitpacker.close(&mut data).unwrap();
assert_eq!(num_bytes, (num_bits * len + 7) / 8);
assert_eq!(data.len(), num_bytes);
let bitunpacker = BitUnpacker::new(&data, num_bits);
bitpacker.close(&mut data).unwrap();
assert_eq!(data.len(), (num_bits * len + 7) / 8 + 7);
let bitunpacker = BitUnpacker::new(data, num_bits);
(bitunpacker, vals)
}
fn test_bitpacker_util(len: usize, num_bits: usize) {
let (bitunpacker, vals) = create_fastfield_bitpacker(len, num_bits);
for (i, val) in vals.iter().enumerate() {
assert_eq!(bitunpacker.get(i), *val);
}
}
#[test]
fn test_bitpacker() {
test_bitpacker_util(10, 3);
@@ -157,4 +220,17 @@ mod test {
test_bitpacker_util(6, 14);
test_bitpacker_util(1000, 14);
}
}
#[test]
fn test_bitpacker_range() {
let (bitunpacker, vals) = create_fastfield_bitpacker(100_000, 12);
let buffer_len = 100;
let mut buffer = vec![0u64; buffer_len];
for start in vec![0, 10, 20, 100, 1_000] {
bitunpacker.get_range(start as u32, &mut buffer[..]);
for i in 0..buffer_len {
assert_eq!(buffer[i], vals[start + i]);
}
}
}
}

184
src/common/composite_file.rs Normal file
View File

@@ -0,0 +1,184 @@
use std::io::Write;
use common::CountingWriter;
use std::collections::HashMap;
use schema::Field;
use common::VInt;
use directory::WritePtr;
use std::io;
use directory::ReadOnlySource;
use common::BinarySerializable;
/// A `CompositeWrite` is used to write a `CompositeFile`.
pub struct CompositeWrite<W = WritePtr> {
write: CountingWriter<W>,
offsets: HashMap<Field, usize>,
}
impl<W: Write> CompositeWrite<W> {
/// Crate a new API writer that writes a composite file
/// in a given write.
pub fn wrap(w: W) -> CompositeWrite<W> {
CompositeWrite {
write: CountingWriter::wrap(w),
offsets: HashMap::new(),
}
}
/// Start writing a new field.
pub fn for_field(&mut self, field: Field) -> &mut CountingWriter<W> {
let offset = self.write.written_bytes();
assert!(!self.offsets.contains_key(&field));
self.offsets.insert(field, offset);
&mut self.write
}
/// Close the composite file.
///
/// An index of the different field offsets
/// will be written as a footer.
pub fn close(mut self) -> io::Result<()> {
let footer_offset = self.write.written_bytes();
VInt(self.offsets.len() as u64).serialize(&mut self.write)?;
let mut offset_fields: Vec<_> = self.offsets
.iter()
.map(|(field, offset)| (offset, field))
.collect();
offset_fields.sort();
let mut prev_offset = 0;
for (offset, field) in offset_fields {
VInt((offset - prev_offset) as u64).serialize(&mut self.write)?;
field.serialize(&mut self.write)?;
prev_offset = *offset;
}
let footer_len = (self.write.written_bytes() - footer_offset) as u32;
footer_len.serialize(&mut self.write)?;
self.write.flush()?;
Ok(())
}
}
/// A composite file is an abstraction to store a
/// file partitioned by field.
///
/// The file needs to be written field by field.
/// A footer describes the start and stop offsets
/// for each field.
#[derive(Clone)]
pub struct CompositeFile {
data: ReadOnlySource,
offsets_index: HashMap<Field, (usize, usize)>,
}
impl CompositeFile {
/// Opens a composite file stored in a given
/// `ReadOnlySource`.
pub fn open(data: &ReadOnlySource) -> io::Result<CompositeFile> {
let end = data.len();
let footer_len_data = data.slice_from(end - 4);
let footer_len = u32::deserialize(&mut footer_len_data.as_slice())? as usize;
let footer_start = end - 4 - footer_len;
let footer_data = data.slice(footer_start, footer_start + footer_len);
let mut footer_buffer = footer_data.as_slice();
let num_fields = VInt::deserialize(&mut footer_buffer)?.0 as usize;
let mut fields = vec![];
let mut offsets = vec![];
let mut field_index = HashMap::new();
let mut offset = 0;
for _ in 0..num_fields {
offset += VInt::deserialize(&mut footer_buffer)?.0 as usize;
let field = Field::deserialize(&mut footer_buffer)?;
offsets.push(offset);
fields.push(field);
}
offsets.push(footer_start);
for i in 0..num_fields {
let field = fields[i];
let start_offset = offsets[i];
let end_offset = offsets[i + 1];
field_index.insert(field, (start_offset, end_offset));
}
Ok(CompositeFile {
data: data.slice_to(footer_start),
offsets_index: field_index,
})
}
/// Returns a composite file that stores
/// no fields.
pub fn empty() -> CompositeFile {
CompositeFile {
offsets_index: HashMap::new(),
data: ReadOnlySource::empty(),
}
}
/// Returns the `ReadOnlySource` associated
/// to a given `Field` and stored in a `CompositeFile`.
pub fn open_read(&self, field: Field) -> Option<ReadOnlySource> {
self.offsets_index
.get(&field)
.map(|&(from, to)| self.data.slice(from, to))
}
}
#[cfg(test)]
mod test {
use std::io::Write;
use super::{CompositeFile, CompositeWrite};
use directory::{Directory, RAMDirectory};
use schema::Field;
use common::VInt;
use common::BinarySerializable;
use std::path::Path;
#[test]
fn test_composite_file() {
let path = Path::new("test_path");
let mut directory = RAMDirectory::create();
{
let w = directory.open_write(path).unwrap();
let mut composite_write = CompositeWrite::wrap(w);
{
let mut write_0 = composite_write.for_field(Field(0u32));
VInt(32431123u64).serialize(&mut write_0).unwrap();
write_0.flush().unwrap();
}
{
let mut write_4 = composite_write.for_field(Field(4u32));
VInt(2).serialize(&mut write_4).unwrap();
write_4.flush().unwrap();
}
composite_write.close().unwrap();
}
{
let r = directory.open_read(path).unwrap();
let composite_file = CompositeFile::open(&r).unwrap();
{
let file0 = composite_file.open_read(Field(0u32)).unwrap();
let mut file0_buf = file0.as_slice();
let payload_0 = VInt::deserialize(&mut file0_buf).unwrap().0;
assert_eq!(file0_buf.len(), 0);
assert_eq!(payload_0, 32431123u64);
}
{
let file4 = composite_file.open_read(Field(4u32)).unwrap();
let mut file4_buf = file4.as_slice();
let payload_4 = VInt::deserialize(&mut file4_buf).unwrap().0;
assert_eq!(file4_buf.len(), 0);
assert_eq!(payload_4, 2u64);
}
}
}
}

55
src/common/counting_writer.rs Normal file
View File

@@ -0,0 +1,55 @@
use std::io::Write;
use std::io;
pub struct CountingWriter<W> {
underlying: W,
written_bytes: usize,
}
impl<W: Write> CountingWriter<W> {
pub fn wrap(underlying: W) -> CountingWriter<W> {
CountingWriter {
underlying,
written_bytes: 0,
}
}
pub fn written_bytes(&self) -> usize {
self.written_bytes
}
pub fn finish(mut self) -> io::Result<(W, usize)> {
self.flush()?;
Ok((self.underlying, self.written_bytes))
}
}
impl<W: Write> Write for CountingWriter<W> {
fn write(&mut self, buf: &[u8]) -> io::Result<usize> {
let written_size = self.underlying.write(buf)?;
self.written_bytes += written_size;
Ok(written_size)
}
fn flush(&mut self) -> io::Result<()> {
self.underlying.flush()
}
}
#[cfg(test)]
mod test {
use super::CountingWriter;
use std::io::Write;
#[test]
fn test_counting_writer() {
let buffer: Vec<u8> = vec![];
let mut counting_writer = CountingWriter::wrap(buffer);
let bytes = (0u8..10u8).collect::<Vec<u8>>();
counting_writer.write_all(&bytes).unwrap();
let (w, len): (Vec<u8>, usize) = counting_writer.finish().unwrap();
assert_eq!(len, 10);
assert_eq!(w.len(), 10);
}
}

67
src/common/mod.rs
View File

@@ -1,32 +1,87 @@
mod serialize;
mod timer;
mod vint;
mod counting_writer;
mod composite_file;
pub mod bitpacker;
pub(crate) use self::composite_file::{CompositeFile, CompositeWrite};
pub use self::serialize::BinarySerializable;
pub use self::timer::Timing;
pub use self::timer::TimerTree;
pub use self::timer::OpenTimer;
pub use self::vint::VInt;
pub use self::counting_writer::CountingWriter;
use std::io;
/// Create a default io error given a string.
pub fn make_io_err(msg: String) -> io::Error {
io::Error::new(io::ErrorKind::Other, msg)
}
/// Has length trait
pub trait HasLen {
/// Return length
fn len(&self,) -> usize;
fn len(&self) -> usize;
/// Returns true iff empty.
fn is_empty(&self,) -> bool {
fn is_empty(&self) -> bool {
self.len() == 0
}
}
const HIGHEST_BIT: u64 = 1 << 63;
/// Maps a `i64` to `u64`
///
/// For simplicity, tantivy internally handles `i64` as `u64`.
/// The mapping is defined by this function.
///
/// Maps `i64` to `u64` so that
/// `-2^63 .. 2^63-1` is mapped
/// to
/// `0 .. 2^64-1`
/// in that order.
///
/// This is more suited than simply casting (`val as u64`)
/// because of bitpacking.
///
/// Imagine a list of `i64` ranging from -10 to 10.
/// When casting negative values, the negative values are projected
/// to values over 2^63, and all values end up requiring 64 bits.
///
/// # See also
/// The [reverse mapping is `u64_to_i64`](./fn.u64_to_i64.html).
#[inline(always)]
pub fn i64_to_u64(val: i64) -> u64 {
(val as u64) ^ HIGHEST_BIT
}
/// Reverse the mapping given by [`i64_to_u64`](./fn.i64_to_u64.html).
#[inline(always)]
pub fn u64_to_i64(val: u64) -> i64 {
(val ^ HIGHEST_BIT) as i64
}
#[cfg(test)]
mod test {
use super::{i64_to_u64, u64_to_i64};
fn test_i64_converter_helper(val: i64) {
assert_eq!(u64_to_i64(i64_to_u64(val)), val);
}
#[test]
fn test_i64_converter() {
assert_eq!(i64_to_u64(i64::min_value()), u64::min_value());
assert_eq!(i64_to_u64(i64::max_value()), u64::max_value());
test_i64_converter_helper(0i64);
test_i64_converter_helper(i64::min_value());
test_i64_converter_helper(i64::max_value());
for i in -1000i64..1000i64 {
test_i64_converter_helper(i);
}
}
}

106
src/common/serialize.rs
View File

@@ -1,4 +1,3 @@
use byteorder::{ReadBytesExt, WriteBytesExt};
use byteorder::LittleEndian as Endianness;
use std::fmt;
@@ -6,115 +5,104 @@ use std::io::Write;
use std::io::Read;
use std::io;
use common::VInt;
use byteorder;
pub trait BinarySerializable : fmt::Debug + Sized {
fn serialize(&self, writer: &mut Write) -> io::Result<usize>;
fn deserialize(reader: &mut Read) -> io::Result<Self>;
}
fn convert_byte_order_error(byteorder_error: byteorder::Error) -> io::Error {
match byteorder_error {
byteorder::Error::UnexpectedEOF => io::Error::new(io::ErrorKind::InvalidData, "Reached EOF unexpectedly"),
byteorder::Error::Io(e) => e,
}
pub trait BinarySerializable: fmt::Debug + Sized {
fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()>;
fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self>;
}
impl BinarySerializable for () {
fn serialize(&self, _: &mut Write) -> io::Result<usize> {
Ok(0)
fn serialize<W: Write>(&self, _: &mut W) -> io::Result<()> {
Ok(())
}
fn deserialize(_: &mut Read) -> io::Result<Self> {
fn deserialize<R: Read>(_: &mut R) -> io::Result<Self> {
Ok(())
}
}
impl<T: BinarySerializable> BinarySerializable for Vec<T> {
fn serialize(&self, writer: &mut Write) -> io::Result<usize> {
let mut total_size = try!(VInt(self.len() as u64).serialize(writer));
fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
VInt(self.len() as u64).serialize(writer)?;
for it in self {
total_size += try!(it.serialize(writer));
it.serialize(writer)?;
}
Ok(total_size)
Ok(())
}
fn deserialize(reader: &mut Read) -> io::Result<Vec<T>> {
let num_items = try!(VInt::deserialize(reader)).val();
fn deserialize<R: Read>(reader: &mut R) -> io::Result<Vec<T>> {
let num_items = VInt::deserialize(reader)?.val();
let mut items: Vec<T> = Vec::with_capacity(num_items as usize);
for _ in 0..num_items {
let item = try!(T::deserialize(reader));
let item = T::deserialize(reader)?;
items.push(item);
}
Ok(items)
}
}
impl<Left: BinarySerializable, Right: BinarySerializable> BinarySerializable for (Left, Right) {
fn serialize(&self, write: &mut Write) -> io::Result<usize> {
Ok(try!(self.0.serialize(write)) + try!(self.1.serialize(write)))
fn serialize<W: Write>(&self, write: &mut W) -> io::Result<()> {
self.0.serialize(write)?;
self.1.serialize(write)
}
fn deserialize(reader: &mut Read) -> io::Result<Self> {
Ok( (try!(Left::deserialize(reader)), try!(Right::deserialize(reader))) )
fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
Ok((Left::deserialize(reader)?, Right::deserialize(reader)?))
}
}
impl BinarySerializable for u32 {
fn serialize(&self, writer: &mut Write) -> io::Result<usize> {
fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
writer.write_u32::<Endianness>(*self)
.map(|_| 4)
.map_err(convert_byte_order_error)
}
fn deserialize(reader: &mut Read) -> io::Result<u32> {
fn deserialize<R: Read>(reader: &mut R) -> io::Result<u32> {
reader.read_u32::<Endianness>()
.map_err(convert_byte_order_error)
}
}
impl BinarySerializable for u64 {
fn serialize(&self, writer: &mut Write) -> io::Result<usize> {
fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
writer.write_u64::<Endianness>(*self)
.map(|_| 8)
.map_err(convert_byte_order_error)
}
fn deserialize(reader: &mut Read) -> io::Result<u64> {
fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
reader.read_u64::<Endianness>()
.map_err(convert_byte_order_error)
}
}
impl BinarySerializable for i64 {
fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
writer.write_i64::<Endianness>(*self)
}
fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
reader.read_i64::<Endianness>()
}
}
impl BinarySerializable for u8 {
fn serialize(&self, writer: &mut Write) -> io::Result<usize> {
// TODO error
try!(writer.write_u8(*self).map_err(convert_byte_order_error));
Ok(1)
fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
writer.write_u8(*self)
}
fn deserialize(reader: &mut Read) -> io::Result<u8> {
fn deserialize<R: Read>(reader: &mut R) -> io::Result<u8> {
reader.read_u8()
.map_err(convert_byte_order_error)
}
}
impl BinarySerializable for String {
fn serialize(&self, writer: &mut Write) -> io::Result<usize> {
fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
let data: &[u8] = self.as_bytes();
let mut size = try!(VInt(data.len() as u64).serialize(writer));
size += data.len();
try!(writer.write_all(data));
Ok(size)
VInt(data.len() as u64).serialize(writer)?;
writer.write_all(data)
}
fn deserialize(reader: &mut Read) -> io::Result<String> {
let string_length = try!(VInt::deserialize(reader)).val() as usize;
fn deserialize<R: Read>(reader: &mut R) -> io::Result<String> {
let string_length = VInt::deserialize(reader)?.val() as usize;
let mut result = String::with_capacity(string_length);
try!(reader.take(string_length as u64).read_to_string(&mut result));
reader
.take(string_length as u64)
.read_to_string(&mut result)?;
Ok(result)
}
}
#[cfg(test)]
mod test {
@@ -123,12 +111,10 @@ mod test {
fn serialize_test<T: BinarySerializable + Eq>(v: T, num_bytes: usize) {
let mut buffer: Vec<u8> = Vec::new();
if num_bytes != 0 {
assert_eq!(v.serialize(&mut buffer).unwrap(), num_bytes);
v.serialize(&mut buffer).unwrap();
assert_eq!(buffer.len(), num_bytes);
}
else {
} else {
v.serialize(&mut buffer).unwrap();
}
let mut cursor = &buffer[..];
@@ -152,15 +138,15 @@ mod test {
#[test]
fn test_serialize_string() {
serialize_test(String::from(""), 1);
serialize_test(String::from("ぽよぽよ"), 1 + 3*4);
serialize_test(String::from("富士さん見える。"), 1 + 3*8);
serialize_test(String::from("ぽよぽよ"), 1 + 3 * 4);
serialize_test(String::from("富士さん見える。"), 1 + 3 * 8);
}
#[test]
fn test_serialize_vec() {
let v: Vec<u8> = Vec::new();
serialize_test(v, 1);
serialize_test(vec!(1u32, 3u32), 1 + 4*2);
serialize_test(vec![1u32, 3u32], 1 + 4 * 2);
}
#[test]

23
src/common/timer.rs
View File

@@ -10,7 +10,7 @@ pub struct OpenTimer<'a> {
impl<'a> OpenTimer<'a> {
/// Starts timing a new named subtask
///
/// The timer is stopped automatically
/// The timer is stopped automatically
/// when the `OpenTimer` is dropped.
pub fn open(&mut self, name: &'static str) -> OpenTimer {
OpenTimer {
@@ -23,17 +23,20 @@ impl<'a> OpenTimer<'a> {
}
impl<'a> Drop for OpenTimer<'a> {
fn drop(&mut self,) {
self.timer_tree.timings.push(Timing {
fn drop(&mut self) {
self.timer_tree.timings.push(Timing {
name: self.name,
duration: self.start.to(PreciseTime::now()).num_microseconds().unwrap(),
duration: self.start
.to(PreciseTime::now())
.num_microseconds()
.unwrap(),
depth: self.depth,
});
}
}
/// Timing recording
#[derive(Debug, RustcEncodable)]
#[derive(Debug, Serialize)]
pub struct Timing {
name: &'static str,
duration: i64,
@@ -41,18 +44,17 @@ pub struct Timing {
}
/// Timer tree
#[derive(Debug, RustcEncodable)]
#[derive(Debug, Serialize)]
pub struct TimerTree {
timings: Vec<Timing>,
}
impl TimerTree {
/// Returns the total time elapsed in microseconds
pub fn total_time(&self,) -> i64 {
/// Returns the total time elapsed in microseconds
pub fn total_time(&self) -> i64 {
self.timings.last().unwrap().duration
}
/// Open a new named subtask
pub fn open(&mut self, name: &'static str) -> OpenTimer {
OpenTimer {
@@ -72,7 +74,6 @@ impl Default for TimerTree {
}
}
#[cfg(test)]
mod tests {

34
src/common/vint.rs
View File

@@ -3,59 +3,55 @@ use std::io;
use std::io::Write;
use std::io::Read;
/// Wrapper over a `u64` that serializes as a variable int.
/// Wrapper over a `u64` that serializes as a variable int.
#[derive(Debug, Eq, PartialEq)]
pub struct VInt(pub u64);
impl VInt {
pub fn val(&self,) -> u64 {
pub fn val(&self) -> u64 {
self.0
}
}
impl BinarySerializable for VInt {
fn serialize(&self, writer: &mut Write) -> io::Result<usize> {
fn serialize<W: Write>(&self, writer: &mut W) -> io::Result<()> {
let mut remaining = self.0;
let mut written: usize = 0;
let mut buffer = [0u8; 10];
let mut i = 0;
loop {
let next_byte: u8 = (remaining % 128u64) as u8;
remaining /= 128u64;
if remaining == 0u64 {
buffer[written] = next_byte | 128u8;
written += 1;
break;
}
else {
buffer[written] = next_byte;
written += 1;
buffer[i] = next_byte | 128u8;
return writer.write_all(&buffer[0..i + 1]);
} else {
buffer[i] = next_byte;
}
i += 1;
}
try!(writer.write_all(&buffer[0..written]));
Ok(written)
}
fn deserialize(reader: &mut Read) -> io::Result<Self> {
fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
let mut bytes = reader.bytes();
let mut result = 0u64;
let mut shift = 0u64;
loop {
match bytes.next() {
Some(Ok(b)) => {
result += ((b % 128u8) as u64) << shift;
result += u64::from(b % 128u8) << shift;
if b & 128u8 != 0u8 {
break;
}
shift += 7;
}
_ => {
return Err(io::Error::new(io::ErrorKind::InvalidData, "Reach end of buffer"))
return Err(io::Error::new(
io::ErrorKind::InvalidData,
"Reach end of buffer",
))
}
}
}
Ok(VInt(result))
}
}

159
src/compression/composite.rs
View File

@@ -1,159 +0,0 @@
use super::{BlockEncoder, BlockDecoder};
use super::NUM_DOCS_PER_BLOCK;
use compression::{VIntEncoder, VIntDecoder};
pub struct CompositeEncoder {
block_encoder: BlockEncoder,
output: Vec<u8>,
}
impl CompositeEncoder {
pub fn new() -> CompositeEncoder {
CompositeEncoder {
block_encoder: BlockEncoder::new(),
output: Vec::with_capacity(500_000),
}
}
pub fn compress_sorted(&mut self, vals: &[u32]) -> &[u8] {
self.output.clear();
let num_blocks = vals.len() / NUM_DOCS_PER_BLOCK;
let mut offset = 0u32;
for i in 0..num_blocks {
let vals_slice = &vals[i * NUM_DOCS_PER_BLOCK .. (i + 1) * NUM_DOCS_PER_BLOCK];
let block_compressed = self.block_encoder.compress_block_sorted(vals_slice, offset);
offset = vals_slice[NUM_DOCS_PER_BLOCK - 1];
self.output.extend_from_slice(block_compressed);
}
let vint_compressed = self.block_encoder.compress_vint_sorted(&vals[num_blocks * NUM_DOCS_PER_BLOCK..], offset);
self.output.extend_from_slice(vint_compressed);
&self.output
}
pub fn compress_unsorted(&mut self, vals: &[u32]) -> &[u8] {
self.output.clear();
let num_blocks = vals.len() / NUM_DOCS_PER_BLOCK;
for i in 0..num_blocks {
let vals_slice = &vals[i * NUM_DOCS_PER_BLOCK .. (i + 1) * NUM_DOCS_PER_BLOCK];
let block_compressed = self.block_encoder.compress_block_unsorted(vals_slice);
self.output.extend_from_slice(block_compressed);
}
let vint_compressed = self.block_encoder.compress_vint_unsorted(&vals[num_blocks * NUM_DOCS_PER_BLOCK..]);
self.output.extend_from_slice(vint_compressed);
&self.output
}
}
pub struct CompositeDecoder {
block_decoder: BlockDecoder,
vals: Vec<u32>,
}
impl CompositeDecoder {
pub fn new() -> CompositeDecoder {
CompositeDecoder {
block_decoder: BlockDecoder::new(),
vals: Vec::with_capacity(500_000),
}
}
pub fn uncompress_sorted(&mut self, mut compressed_data: &[u8], uncompressed_len: usize) -> &[u32] {
if uncompressed_len > self.vals.capacity() {
let extra_capacity = uncompressed_len - self.vals.capacity();
self.vals.reserve(extra_capacity);
}
let mut offset = 0u32;
self.vals.clear();
let num_blocks = uncompressed_len / NUM_DOCS_PER_BLOCK;
for _ in 0..num_blocks {
compressed_data = self.block_decoder.uncompress_block_sorted(compressed_data, offset);
offset = self.block_decoder.output(NUM_DOCS_PER_BLOCK - 1);
self.vals.extend_from_slice(self.block_decoder.output_array());
}
self.block_decoder.uncompress_vint_sorted(compressed_data, offset, uncompressed_len % NUM_DOCS_PER_BLOCK);
self.vals.extend_from_slice(self.block_decoder.output_array());
&self.vals
}
pub fn uncompress_unsorted(&mut self, mut compressed_data: &[u8], uncompressed_len: usize) -> &[u32] {
self.vals.clear();
let num_blocks = uncompressed_len / NUM_DOCS_PER_BLOCK;
for _ in 0..num_blocks {
compressed_data = self.block_decoder.uncompress_block_unsorted(compressed_data);
self.vals.extend_from_slice(self.block_decoder.output_array());
}
self.block_decoder.uncompress_vint_unsorted(compressed_data, uncompressed_len % NUM_DOCS_PER_BLOCK);
self.vals.extend_from_slice(self.block_decoder.output_array());
&self.vals
}
}
impl Into<Vec<u32>> for CompositeDecoder {
fn into(self) -> Vec<u32> {
self.vals
}
}
#[cfg(test)]
pub mod tests {
use test::Bencher;
use super::*;
use compression::tests::generate_array;
#[test]
fn test_composite_unsorted() {
let data = generate_array(10_000, 0.1);
let mut encoder = CompositeEncoder::new();
let compressed = encoder.compress_unsorted(&data);
assert_eq!(compressed.len(), 19_790);
let mut decoder = CompositeDecoder::new();
let result = decoder.uncompress_unsorted(&compressed, data.len());
for i in 0..data.len() {
assert_eq!(data[i], result[i]);
}
}
#[test]
fn test_composite_sorted() {
let data = generate_array(10_000, 0.1);
let mut encoder = CompositeEncoder::new();
let compressed = encoder.compress_sorted(&data);
assert_eq!(compressed.len(), 7_822);
let mut decoder = CompositeDecoder::new();
let result = decoder.uncompress_sorted(&compressed, data.len());
for i in 0..data.len() {
assert_eq!(data[i], result[i]);
}
}
const BENCH_NUM_INTS: usize = 99_968;
#[bench]
fn bench_compress(b: &mut Bencher) {
let mut encoder = CompositeEncoder::new();
let data = generate_array(BENCH_NUM_INTS, 0.1);
b.iter(|| {
encoder.compress_sorted(&data);
});
}
#[bench]
fn bench_uncompress(b: &mut Bencher) {
let mut encoder = CompositeEncoder::new();
let data = generate_array(BENCH_NUM_INTS, 0.1);
let compressed = encoder.compress_sorted(&data);
let mut decoder = CompositeDecoder::new();
b.iter(|| {
decoder.uncompress_sorted(compressed, BENCH_NUM_INTS);
});
}
}

127
src/compression/compression_nosimd.rs
View File

@@ -1,127 +0,0 @@
use common::bitpacker::compute_num_bits;
use common::bitpacker::{BitPacker, BitUnpacker};
use std::cmp;
use std::io::Write;
use super::NUM_DOCS_PER_BLOCK;
const COMPRESSED_BLOCK_MAX_SIZE: usize = NUM_DOCS_PER_BLOCK * 4 + 1;
pub fn compress_sorted(vals: &mut [u32], mut output: &mut [u8], offset: u32) -> usize {
let mut max_delta = 0;
{
let mut local_offset = offset;
for i in 0..NUM_DOCS_PER_BLOCK {
let val = vals[i];
let delta = val - local_offset;
max_delta = cmp::max(max_delta, delta);
vals[i] = delta;
local_offset = val;
}
}
let num_bits = compute_num_bits(max_delta);
output.write_all(&[num_bits]).unwrap();
let mut bit_packer = BitPacker::new(num_bits as usize);
for val in vals {
bit_packer.write(*val, &mut output).unwrap();
}
1 + bit_packer.close(&mut output).expect("packing in memory should never fail")
}
pub struct BlockEncoder {
pub output: [u8; COMPRESSED_BLOCK_MAX_SIZE],
pub output_len: usize,
input_buffer: [u32; NUM_DOCS_PER_BLOCK],
}
impl BlockEncoder {
pub fn new() -> BlockEncoder {
BlockEncoder {
output: [0u8; COMPRESSED_BLOCK_MAX_SIZE],
output_len: 0,
input_buffer: [0u32; NUM_DOCS_PER_BLOCK],
}
}
pub fn compress_block_sorted(&mut self, vals: &[u32], offset: u32) -> &[u8] {
self.input_buffer.clone_from_slice(vals);
let compressed_size = compress_sorted(&mut self.input_buffer, &mut self.output, offset);
&self.output[..compressed_size]
}
pub fn compress_block_unsorted(&mut self, vals: &[u32]) -> &[u8] {
let compressed_size: usize = {
let mut output: &mut [u8] = &mut self.output;
let max = vals.iter().cloned().max().expect("compress unsorted called with an empty array");
let num_bits = compute_num_bits(max);
output.write_all(&[num_bits]).unwrap();
let mut bit_packer = BitPacker::new(num_bits as usize);
for val in vals {
bit_packer.write(*val, &mut output).unwrap();
}
1 + bit_packer.close(&mut output).expect("packing in memory should never fail")
};
&self.output[..compressed_size]
}
}
pub struct BlockDecoder {
pub output: [u32; COMPRESSED_BLOCK_MAX_SIZE],
pub output_len: usize,
}
impl BlockDecoder {
pub fn new() -> BlockDecoder {
BlockDecoder::with_val(0u32)
}
pub fn with_val(val: u32) -> BlockDecoder {
BlockDecoder {
output: [val; COMPRESSED_BLOCK_MAX_SIZE],
output_len: 0,
}
}
pub fn uncompress_block_sorted<'a>(&mut self, compressed_data: &'a [u8], mut offset: u32) -> &'a[u8] {
let consumed_size = {
let num_bits = compressed_data[0];
let bit_unpacker = BitUnpacker::new(&compressed_data[1..], num_bits as usize);
for i in 0..NUM_DOCS_PER_BLOCK {
let delta = bit_unpacker.get(i);
let val = offset + delta;
self.output[i] = val;
offset = val;
}
1 + (num_bits as usize * NUM_DOCS_PER_BLOCK + 7) / 8
};
self.output_len = NUM_DOCS_PER_BLOCK;
&compressed_data[consumed_size..]
}
pub fn uncompress_block_unsorted<'a>(&mut self, compressed_data: &'a [u8]) -> &'a[u8] {
let num_bits = compressed_data[0];
let bit_unpacker = BitUnpacker::new(&compressed_data[1..], num_bits as usize);
for i in 0..NUM_DOCS_PER_BLOCK {
self.output[i] = bit_unpacker.get(i);
}
let consumed_size = 1 + (num_bits as usize * NUM_DOCS_PER_BLOCK + 7) / 8;
self.output_len = NUM_DOCS_PER_BLOCK;
&compressed_data[consumed_size..]
}
#[inline]
pub fn output_array(&self,) -> &[u32] {
&self.output[..self.output_len]
}
#[inline]
pub fn output(&self, idx: usize) -> u32 {
self.output[idx]
}
}

321
src/compression/mod.rs
View File

@@ -1,164 +1,135 @@
#![allow(dead_code)]
mod composite;
pub use self::composite::{CompositeEncoder, CompositeDecoder};
mod stream;
#[cfg(feature="simdcompression")]
mod compression_simd;
#[cfg(feature="simdcompression")]
pub use self::compression_simd::{BlockEncoder, BlockDecoder};
pub use self::stream::CompressedIntStream;
#[cfg(not(feature = "simdcompression"))]
mod pack {
mod compression_pack_nosimd;
pub use self::compression_pack_nosimd::{BlockDecoder, BlockEncoder};
}
#[cfg(not(feature="simdcompression"))]
mod compression_nosimd;
#[cfg(not(feature="simdcompression"))]
pub use self::compression_nosimd::{BlockEncoder, BlockDecoder};
#[cfg(feature = "simdcompression")]
mod pack {
mod compression_pack_simd;
pub use self::compression_pack_simd::{BlockDecoder, BlockEncoder};
}
pub use self::pack::{BlockDecoder, BlockEncoder};
#[cfg(any(not(feature = "simdcompression"), target_env = "msvc"))]
mod vint {
mod compression_vint_nosimd;
pub(crate) use self::compression_vint_nosimd::*;
}
#[cfg(all(feature = "simdcompression", not(target_env = "msvc")))]
mod vint {
mod compression_vint_simd;
pub(crate) use self::compression_vint_simd::*;
}
/// Returns the size in bytes of a compressed block, given `num_bits`.
pub fn compressed_block_size(num_bits: u8) -> usize {
1 + (num_bits as usize) * 16
}
pub trait VIntEncoder {
/// Compresses an array of `u32` integers,
/// using [delta-encoding](https://en.wikipedia.org/wiki/Delta_encoding)
/// and variable bytes encoding.
///
/// The method takes an array of ints to compress, and returns
/// a `&[u8]` representing the compressed data.
///
/// The method also takes an offset to give the value of the
/// hypothetical previous element in the delta-encoding.
fn compress_vint_sorted(&mut self, input: &[u32], offset: u32) -> &[u8];
/// Compresses an array of `u32` integers,
/// using variable bytes encoding.
///
/// The method takes an array of ints to compress, and returns
/// a `&[u8]` representing the compressed data.
fn compress_vint_unsorted(&mut self, input: &[u32]) -> &[u8];
}
pub trait VIntDecoder {
fn uncompress_vint_sorted<'a>(&mut self, compressed_data: &'a [u8], offset: u32, num_els: usize) -> &'a [u8];
fn uncompress_vint_unsorted<'a>(&mut self, compressed_data: &'a [u8], num_els: usize) -> &'a [u8];
}
impl VIntEncoder for BlockEncoder{
fn compress_vint_sorted(&mut self, input: &[u32], mut offset: u32) -> &[u8] {
let mut byte_written = 0;
for &v in input {
let mut to_encode: u32 = v - offset;
offset = v;
loop {
let next_byte: u8 = (to_encode % 128u32) as u8;
to_encode /= 128u32;
if to_encode == 0u32 {
self.output[byte_written] = next_byte | 128u8;
byte_written += 1;
break;
}
else {
self.output[byte_written] = next_byte;
byte_written += 1;
}
}
}
&self.output[..byte_written]
}
fn compress_vint_unsorted(&mut self, input: &[u32]) -> &[u8] {
let mut byte_written = 0;
for &v in input {
let mut to_encode: u32 = v;
loop {
let next_byte: u8 = (to_encode % 128u32) as u8;
to_encode /= 128u32;
if to_encode == 0u32 {
self.output[byte_written] = next_byte | 128u8;
byte_written += 1;
break;
}
else {
self.output[byte_written] = next_byte;
byte_written += 1;
}
}
}
&self.output[..byte_written]
}
}
impl VIntDecoder for BlockDecoder {
/// Uncompress an array of `u32` integers,
/// that were compressed using [delta-encoding](https://en.wikipedia.org/wiki/Delta_encoding)
/// and variable bytes encoding.
///
/// The method takes a number of int to decompress, and returns
/// the amount of bytes that were read to decompress them.
///
/// The method also takes an offset to give the value of the
/// hypothetical previous element in the delta-encoding.
///
/// For instance, if delta encoded are `1, 3, 9`, and the
/// `offset` is 5, then the output will be:
/// `5 + 1 = 6, 6 + 3= 9, 9 + 9 = 18`
fn uncompress_vint_sorted<'a>(
&mut self,
compressed_data: &'a [u8],
offset: u32,
num_els: usize) -> &'a [u8] {
let mut read_byte = 0;
let mut result = offset;
for i in 0..num_els {
let mut shift = 0u32;
loop {
let cur_byte = compressed_data[read_byte];
read_byte += 1;
result += ((cur_byte % 128u8) as u32) << shift;
if cur_byte & 128u8 != 0u8 {
break;
}
shift += 7;
}
self.output[i] = result;
}
self.output_len = num_els;
&compressed_data[read_byte..]
}
fn uncompress_vint_unsorted<'a>(
&mut self,
compressed_data: &'a [u8],
num_els: usize) -> &'a [u8] {
let mut read_byte = 0;
for i in 0..num_els {
let mut result = 0u32;
let mut shift = 0u32;
loop {
let cur_byte = compressed_data[read_byte];
read_byte += 1;
result += ((cur_byte % 128u8) as u32) << shift;
if cur_byte & 128u8 != 0u8 {
break;
}
shift += 7;
}
self.output[i] = result;
}
self.output_len = num_els;
&compressed_data[read_byte..]
}
num_els: usize,
) -> usize;
/// Uncompress an array of `u32s`, compressed using variable
/// byte encoding.
///
/// The method takes a number of int to decompress, and returns
/// the amount of bytes that were read to decompress them.
fn uncompress_vint_unsorted<'a>(&mut self, compressed_data: &'a [u8], num_els: usize) -> usize;
}
impl VIntEncoder for BlockEncoder {
fn compress_vint_sorted(&mut self, input: &[u32], offset: u32) -> &[u8] {
vint::compress_sorted(input, &mut self.output, offset)
}
pub const NUM_DOCS_PER_BLOCK: usize = 128; //< should be a power of 2 to let the compiler optimize.
fn compress_vint_unsorted(&mut self, input: &[u32]) -> &[u8] {
vint::compress_unsorted(input, &mut self.output)
}
}
impl VIntDecoder for BlockDecoder {
fn uncompress_vint_sorted<'a>(
&mut self,
compressed_data: &'a [u8],
offset: u32,
num_els: usize,
) -> usize {
self.output_len = num_els;
vint::uncompress_sorted(compressed_data, &mut self.output[..num_els], offset)
}
fn uncompress_vint_unsorted<'a>(&mut self, compressed_data: &'a [u8], num_els: usize) -> usize {
self.output_len = num_els;
vint::uncompress_unsorted(compressed_data, &mut self.output[..num_els])
}
}
pub const COMPRESSION_BLOCK_SIZE: usize = 128;
#[cfg(test)]
pub mod tests {
use rand::Rng;
use rand::SeedableRng;
use rand::XorShiftRng;
use super::*;
use tests;
use test::Bencher;
fn generate_array_with_seed(n: usize, ratio: f32, seed_val: u32) -> Vec<u32> {
let seed: &[u32; 4] = &[1, 2, 3, seed_val];
let mut rng: XorShiftRng = XorShiftRng::from_seed(*seed);
(0..u32::max_value())
.filter(|_| rng.next_f32()< ratio)
.take(n)
.collect()
}
pub fn generate_array(n: usize, ratio: f32) -> Vec<u32> {
generate_array_with_seed(n, ratio, 4)
}
#[test]
fn test_encode_sorted_block() {
let vals: Vec<u32> = (0u32..128u32).map(|i| i*7).collect();
let vals: Vec<u32> = (0u32..128u32).map(|i| i * 7).collect();
let mut encoder = BlockEncoder::new();
let compressed_data = encoder.compress_block_sorted(&vals, 0);
let mut decoder = BlockDecoder::new();
{
let remaining_data = decoder.uncompress_block_sorted(compressed_data, 0);
assert_eq!(remaining_data.len(), 0);
let consumed_num_bytes = decoder.uncompress_block_sorted(compressed_data, 0);
assert_eq!(consumed_num_bytes, compressed_data.len());
}
for i in 0..128 {
assert_eq!(vals[i], decoder.output(i));
@@ -167,33 +138,33 @@ pub mod tests {
#[test]
fn test_encode_sorted_block_with_offset() {
let vals: Vec<u32> = (0u32..128u32).map(|i| 11 + i*7).collect();
let vals: Vec<u32> = (0u32..128u32).map(|i| 11 + i * 7).collect();
let mut encoder = BlockEncoder::new();
let compressed_data = encoder.compress_block_sorted(&vals, 10);
let mut decoder = BlockDecoder::new();
{
let remaining_data = decoder.uncompress_block_sorted(compressed_data, 10);
assert_eq!(remaining_data.len(), 0);
let consumed_num_bytes = decoder.uncompress_block_sorted(compressed_data, 10);
assert_eq!(consumed_num_bytes, compressed_data.len());
}
for i in 0..128 {
assert_eq!(vals[i], decoder.output(i));
}
}
#[test]
fn test_encode_sorted_block_with_junk() {
let mut compressed: Vec<u8> = Vec::new();
let n = 128;
let vals: Vec<u32> = (0..n).map(|i| 11u32 + (i as u32)*7u32).collect();
let vals: Vec<u32> = (0..n).map(|i| 11u32 + (i as u32) * 7u32).collect();
let mut encoder = BlockEncoder::new();
let compressed_data = encoder.compress_block_sorted(&vals, 10);
compressed.extend_from_slice(compressed_data);
compressed.push(173u8);
let mut decoder = BlockDecoder::new();
{
let remaining_data = decoder.uncompress_block_sorted(&compressed, 10);
assert_eq!(remaining_data.len(), 1);
assert_eq!(remaining_data[0], 173u8);
let consumed_num_bytes = decoder.uncompress_block_sorted(&compressed, 10);
assert_eq!(consumed_num_bytes, compressed.len() - 1);
assert_eq!(compressed[consumed_num_bytes], 173u8);
}
for i in 0..n {
assert_eq!(vals[i], decoder.output(i));
@@ -204,72 +175,94 @@ pub mod tests {
fn test_encode_unsorted_block_with_junk() {
let mut compressed: Vec<u8> = Vec::new();
let n = 128;
let vals: Vec<u32> = (0..n).map(|i| 11u32 + (i as u32)*7u32 % 12).collect();
let vals: Vec<u32> = (0..n).map(|i| 11u32 + (i as u32) * 7u32 % 12).collect();
let mut encoder = BlockEncoder::new();
let compressed_data = encoder.compress_block_unsorted(&vals);
compressed.extend_from_slice(compressed_data);
compressed.push(173u8);
let mut decoder = BlockDecoder::new();
{
let remaining_data = decoder.uncompress_block_unsorted(&compressed);
assert_eq!(remaining_data.len(), 1);
assert_eq!(remaining_data[0], 173u8);
let consumed_num_bytes = decoder.uncompress_block_unsorted(&compressed);
assert_eq!(consumed_num_bytes + 1, compressed.len());
assert_eq!(compressed[consumed_num_bytes], 173u8);
}
for i in 0..n {
assert_eq!(vals[i], decoder.output(i));
}
}
#[test]
fn test_encode_vint() {
{
let expected_length = 123;
let expected_length = 154;
let mut encoder = BlockEncoder::new();
let input: Vec<u32> = (0u32..123u32)
.map(|i| 4 + i * 7 / 2)
.into_iter()
.collect();
let input: Vec<u32> = (0u32..123u32).map(|i| 4 + i * 7 / 2).into_iter().collect();
for offset in &[0u32, 1u32, 2u32] {
let encoded_data = encoder.compress_vint_sorted(&input, *offset);
assert_eq!(encoded_data.len(), expected_length);
assert!(encoded_data.len() <= expected_length);
let mut decoder = BlockDecoder::new();
let remaining_data = decoder.uncompress_vint_sorted(&encoded_data, *offset, input.len());
assert_eq!(0, remaining_data.len());
let consumed_num_bytes =
decoder.uncompress_vint_sorted(&encoded_data, *offset, input.len());
assert_eq!(consumed_num_bytes, encoded_data.len());
assert_eq!(input, decoder.output_array());
}
}
{
let mut encoder = BlockEncoder::new();
let input = vec!(3u32, 17u32, 187u32);
let encoded_data = encoder.compress_vint_sorted(&input, 0);
assert_eq!(encoded_data.len(), 4);
assert_eq!(encoded_data[0], 3u8 + 128u8);
assert_eq!(encoded_data[1], (17u8 - 3u8) + 128u8);
assert_eq!(encoded_data[2], (187u8 - 17u8 - 128u8));
assert_eq!(encoded_data[3], (1u8 + 128u8));
}
}
#[bench]
fn bench_compress(b: &mut Bencher) {
let mut encoder = BlockEncoder::new();
let data = generate_array(NUM_DOCS_PER_BLOCK, 0.1);
let data = tests::generate_array(COMPRESSION_BLOCK_SIZE, 0.1);
b.iter(|| {
encoder.compress_block_sorted(&data, 0u32);
});
}
#[bench]
fn bench_uncompress(b: &mut Bencher) {
let mut encoder = BlockEncoder::new();
let data = generate_array(NUM_DOCS_PER_BLOCK, 0.1);
let data = tests::generate_array(COMPRESSION_BLOCK_SIZE, 0.1);
let compressed = encoder.compress_block_sorted(&data, 0u32);
let mut decoder = BlockDecoder::new();
let mut decoder = BlockDecoder::new();
b.iter(|| {
decoder.uncompress_block_sorted(compressed, 0u32);
});
}
#[test]
fn test_all_docs_compression_numbits() {
for num_bits in 0..33 {
let mut data = [0u32; 128];
if num_bits > 0 {
data[0] = 1 << (num_bits - 1);
}
let mut encoder = BlockEncoder::new();
let compressed = encoder.compress_block_unsorted(&data);
assert_eq!(compressed[0] as usize, num_bits);
assert_eq!(compressed.len(), compressed_block_size(compressed[0]));
}
}
const NUM_INTS_BENCH_VINT: usize = 10;
#[bench]
fn bench_compress_vint(b: &mut Bencher) {
let mut encoder = BlockEncoder::new();
let data = tests::generate_array(NUM_INTS_BENCH_VINT, 0.001);
b.iter(|| {
encoder.compress_vint_sorted(&data, 0u32);
});
}
#[bench]
fn bench_uncompress_vint(b: &mut Bencher) {
let mut encoder = BlockEncoder::new();
let data = tests::generate_array(NUM_INTS_BENCH_VINT, 0.001);
let compressed = encoder.compress_vint_sorted(&data, 0u32);
let mut decoder = BlockDecoder::new();
b.iter(|| {
decoder.uncompress_vint_sorted(compressed, 0u32, NUM_INTS_BENCH_VINT);
});
}
}

142
src/compression/pack/compression_pack_nosimd.rs Normal file
View File

@@ -0,0 +1,142 @@
use common::bitpacker::compute_num_bits;
use common::bitpacker::{BitPacker, BitUnpacker};
use common::CountingWriter;
use std::cmp;
use std::io::Write;
use super::super::COMPRESSION_BLOCK_SIZE;
const COMPRESSED_BLOCK_MAX_SIZE: usize = COMPRESSION_BLOCK_SIZE * 4 + 1;
pub fn compress_sorted(vals: &mut [u32], output: &mut [u8], offset: u32) -> usize {
let mut max_delta = 0;
{
let mut local_offset = offset;
for i in 0..COMPRESSION_BLOCK_SIZE {
let val = vals[i];
let delta = val - local_offset;
max_delta = cmp::max(max_delta, delta);
vals[i] = delta;
local_offset = val;
}
}
let mut counting_writer = CountingWriter::wrap(output);
let num_bits = compute_num_bits(max_delta as u64);
counting_writer.write_all(&[num_bits]).unwrap();
let mut bit_packer = BitPacker::new(num_bits as usize);
for val in vals {
bit_packer.write(*val as u64, &mut counting_writer).unwrap();
}
counting_writer.written_bytes()
}
pub struct BlockEncoder {
pub output: [u8; COMPRESSED_BLOCK_MAX_SIZE],
pub output_len: usize,
input_buffer: [u32; COMPRESSION_BLOCK_SIZE],
}
impl BlockEncoder {
pub fn new() -> BlockEncoder {
BlockEncoder {
output: [0u8; COMPRESSED_BLOCK_MAX_SIZE],
output_len: 0,
input_buffer: [0u32; COMPRESSION_BLOCK_SIZE],
}
}
pub fn compress_block_sorted(&mut self, vals: &[u32], offset: u32) -> &[u8] {
self.input_buffer.clone_from_slice(vals);
let compressed_size = compress_sorted(&mut self.input_buffer, &mut self.output, offset);
&self.output[..compressed_size]
}
pub fn compress_block_unsorted(&mut self, vals: &[u32]) -> &[u8] {
let compressed_size = {
let output: &mut [u8] = &mut self.output;
let max = vals.iter()
.cloned()
.max()
.expect("compress unsorted called with an empty array");
let num_bits = compute_num_bits(max as u64);
let mut counting_writer = CountingWriter::wrap(output);
counting_writer.write_all(&[num_bits]).unwrap();
let mut bit_packer = BitPacker::new(num_bits as usize);
for val in vals {
bit_packer.write(*val as u64, &mut counting_writer).unwrap();
}
for _ in vals.len()..COMPRESSION_BLOCK_SIZE {
bit_packer
.write(vals[0] as u64, &mut counting_writer)
.unwrap();
}
bit_packer.flush(&mut counting_writer).expect(
"Flushing the bitpacking \
in an in RAM buffer should never fail",
);
// we avoid writing "closing", because we
// do not want 7 bytes of padding here.
counting_writer.written_bytes()
};
&self.output[..compressed_size]
}
}
pub struct BlockDecoder {
pub output: [u32; COMPRESSED_BLOCK_MAX_SIZE],
pub output_len: usize,
}
impl BlockDecoder {
pub fn new() -> BlockDecoder {
BlockDecoder::with_val(0u32)
}
pub fn with_val(val: u32) -> BlockDecoder {
BlockDecoder {
output: [val; COMPRESSED_BLOCK_MAX_SIZE],
output_len: 0,
}
}
pub fn uncompress_block_sorted<'a>(
&mut self,
compressed_data: &'a [u8],
mut offset: u32,
) -> usize {
let consumed_size = {
let num_bits = compressed_data[0];
let bit_unpacker = BitUnpacker::new(&compressed_data[1..], num_bits as usize);
for i in 0..COMPRESSION_BLOCK_SIZE {
let delta = bit_unpacker.get(i);
let val = offset + delta as u32;
self.output[i] = val;
offset = val;
}
1 + (num_bits as usize * COMPRESSION_BLOCK_SIZE + 7) / 8
};
self.output_len = COMPRESSION_BLOCK_SIZE;
consumed_size
}
pub fn uncompress_block_unsorted<'a>(&mut self, compressed_data: &'a [u8]) -> usize {
let num_bits = compressed_data[0];
let bit_unpacker = BitUnpacker::new(&compressed_data[1..], num_bits as usize);
for i in 0..COMPRESSION_BLOCK_SIZE {
self.output[i] = bit_unpacker.get(i) as u32;
}
let consumed_size = 1 + (num_bits as usize * COMPRESSION_BLOCK_SIZE + 7) / 8;
self.output_len = COMPRESSION_BLOCK_SIZE;
consumed_size
}
#[inline]
pub fn output_array(&self) -> &[u32] {
&self.output[..self.output_len]
}
#[inline]
pub fn output(&self, idx: usize) -> u32 {
self.output[idx]
}
}

88
src/compression/compression_simd.rs → src/compression/pack/compression_pack_simd.rs
View File

@@ -1,71 +1,63 @@
use super::super::COMPRESSION_BLOCK_SIZE;
use super::NUM_DOCS_PER_BLOCK;
const COMPRESSED_BLOCK_MAX_SIZE: usize = COMPRESSION_BLOCK_SIZE * 4 + 1;
use libc::size_t;
mod simdcomp {
use libc::size_t;
const COMPRESSED_BLOCK_MAX_SIZE: usize = NUM_DOCS_PER_BLOCK * 4 + 1;
extern "C" {
pub fn compress_sorted(data: *const u32, output: *mut u8, offset: u32) -> size_t;
extern {
fn compress_sorted_cpp(
data: *const u32,
output: *mut u8,
offset: u32) -> size_t;
pub fn uncompress_sorted(
compressed_data: *const u8,
output: *mut u32,
offset: u32,
) -> size_t;
fn uncompress_sorted_cpp(
compressed_data: *const u8,
output: *mut u32,
offset: u32) -> size_t;
fn compress_unsorted_cpp(
data: *const u32,
output: *mut u8) -> size_t;
pub fn compress_unsorted(data: *const u32, output: *mut u8) -> size_t;
fn uncompress_unsorted_cpp(
compressed_data: *const u8,
output: *mut u32) -> size_t;
pub fn uncompress_unsorted(compressed_data: *const u8, output: *mut u32) -> size_t;
}
}
fn compress_sorted(vals: &[u32], output: &mut [u8], offset: u32) -> usize {
unsafe { compress_sorted_cpp(vals.as_ptr(), output.as_mut_ptr(), offset) }
unsafe { simdcomp::compress_sorted(vals.as_ptr(), output.as_mut_ptr(), offset) }
}
fn uncompress_sorted(compressed_data: &[u8], output: &mut [u32], offset: u32) -> usize {
unsafe { uncompress_sorted_cpp(compressed_data.as_ptr(), output.as_mut_ptr(), offset) }
unsafe { simdcomp::uncompress_sorted(compressed_data.as_ptr(), output.as_mut_ptr(), offset) }
}
fn compress_unsorted(vals: &[u32], output: &mut [u8]) -> usize {
unsafe { compress_unsorted_cpp(vals.as_ptr(), output.as_mut_ptr()) }
unsafe { simdcomp::compress_unsorted(vals.as_ptr(), output.as_mut_ptr()) }
}
fn uncompress_unsorted(compressed_data: &[u8], output: &mut [u32]) -> usize {
unsafe { uncompress_unsorted_cpp(compressed_data.as_ptr(), output.as_mut_ptr()) }
unsafe { simdcomp::uncompress_unsorted(compressed_data.as_ptr(), output.as_mut_ptr()) }
}
pub struct BlockEncoder {
pub output: [u8; COMPRESSED_BLOCK_MAX_SIZE],
pub output_len: usize,
}
impl BlockEncoder {
pub fn new() -> BlockEncoder {
BlockEncoder {
output: [0u8; COMPRESSED_BLOCK_MAX_SIZE],
output_len: 0,
}
}
}
pub fn compress_block_sorted(&mut self, vals: &[u32], offset: u32) -> &[u8] {
let compressed_size = compress_sorted(vals, &mut self.output, offset);
&self.output[..compressed_size]
}
pub fn compress_block_unsorted(&mut self, vals: &[u32]) -> &[u8] {
let compressed_size = compress_unsorted(vals, &mut self.output);
&self.output[..compressed_size]
}
}
pub struct BlockDecoder {
@@ -73,40 +65,52 @@ pub struct BlockDecoder {
pub output_len: usize,
}
impl BlockDecoder {
pub fn new() -> BlockDecoder {
BlockDecoder::with_val(0u32)
}
pub fn with_val(val: u32) -> BlockDecoder {
BlockDecoder {
output: [val; COMPRESSED_BLOCK_MAX_SIZE],
output_len: 0,
}
}
pub fn uncompress_block_sorted<'a>(&mut self, compressed_data: &'a [u8], offset: u32) -> &'a[u8] {
pub fn uncompress_block_sorted(&mut self, compressed_data: &[u8], offset: u32) -> usize {
let consumed_size = uncompress_sorted(compressed_data, &mut self.output, offset);
self.output_len = NUM_DOCS_PER_BLOCK;
&compressed_data[consumed_size..]
self.output_len = COMPRESSION_BLOCK_SIZE;
consumed_size
}
pub fn uncompress_block_unsorted<'a>(&mut self, compressed_data: &'a [u8]) -> &'a[u8] {
pub fn uncompress_block_unsorted<'a>(&mut self, compressed_data: &'a [u8]) -> usize {
let consumed_size = uncompress_unsorted(compressed_data, &mut self.output);
self.output_len = NUM_DOCS_PER_BLOCK;
&compressed_data[consumed_size..]
self.output_len = COMPRESSION_BLOCK_SIZE;
consumed_size
}
#[inline]
pub fn output_array(&self,) -> &[u32] {
pub fn output_array(&self) -> &[u32] {
&self.output[..self.output_len]
}
#[inline]
pub fn output(&self, idx: usize) -> u32 {
self.output[idx]
}
}
#[cfg(test)]
mod tests {
use super::BlockEncoder;
#[test]
fn test_all_docs_compression_len() {
let data: Vec<u32> = (0u32..128u32).collect();
let mut encoder = BlockEncoder::new();
let compressed = encoder.compress_block_sorted(&data, 0u32);
assert_eq!(compressed.len(), 17);
}
}

130
src/compression/stream.rs Normal file
View File

@@ -0,0 +1,130 @@
use compression::BlockDecoder;
use compression::COMPRESSION_BLOCK_SIZE;
use compression::compressed_block_size;
use directory::{ReadOnlySource, SourceRead};
/// Reads a stream of compressed ints.
///
/// Tantivy uses `CompressedIntStream` to read
/// the position file.
/// The `.skip(...)` makes it possible to avoid
/// decompressing blocks that are not required.
pub struct CompressedIntStream {
buffer: SourceRead,
block_decoder: BlockDecoder,
inner_offset: usize,
}
impl CompressedIntStream {
/// Opens a compressed int stream.
pub(crate) fn wrap(source: ReadOnlySource) -> CompressedIntStream {
CompressedIntStream {
buffer: SourceRead::from(source),
block_decoder: BlockDecoder::new(),
inner_offset: COMPRESSION_BLOCK_SIZE,
}
}
/// Fills a buffer with the next `output.len()` integers,
/// and advance the stream by that many els.
pub fn read(&mut self, output: &mut [u32]) {
let mut num_els: usize = output.len();
let mut start: usize = 0;
loop {
let available = COMPRESSION_BLOCK_SIZE - self.inner_offset;
if num_els >= available {
if available > 0 {
let uncompressed_block =
&self.block_decoder.output_array()[self.inner_offset..];
output[start..][..available].clone_from_slice(uncompressed_block);
}
num_els -= available;
start += available;
let num_consumed_bytes = self.block_decoder
.uncompress_block_unsorted(self.buffer.as_ref());
self.buffer.advance(num_consumed_bytes);
self.inner_offset = 0;
} else {
let uncompressed_block = &self.block_decoder.output_array()
[self.inner_offset..self.inner_offset + num_els];
output[start..][..num_els].clone_from_slice(uncompressed_block);
self.inner_offset += num_els;
break;
}
}
}
/// Skip the next `skip_len` integer.
///
/// If a full block is skipped, calling
/// `.skip(...)` will avoid decompressing it.
pub fn skip(&mut self, mut skip_len: usize) {
let available = COMPRESSION_BLOCK_SIZE - self.inner_offset;
if available >= skip_len {
self.inner_offset += skip_len;
} else {
skip_len -= available;
// entirely skip decompressing some blocks.
while skip_len >= COMPRESSION_BLOCK_SIZE {
skip_len -= COMPRESSION_BLOCK_SIZE;
let num_bits: u8 = self.buffer.as_ref()[0];
let block_len = compressed_block_size(num_bits);
self.buffer.advance(block_len);
}
let num_consumed_bytes = self.block_decoder
.uncompress_block_unsorted(self.buffer.as_ref());
self.buffer.advance(num_consumed_bytes);
self.inner_offset = skip_len;
}
}
}
#[cfg(test)]
pub mod tests {
use super::CompressedIntStream;
use compression::compressed_block_size;
use compression::COMPRESSION_BLOCK_SIZE;
use compression::BlockEncoder;
use directory::ReadOnlySource;
fn create_stream_buffer() -> ReadOnlySource {
let mut buffer: Vec<u8> = vec![];
let mut encoder = BlockEncoder::new();
let vals: Vec<u32> = (0u32..1_025u32).collect();
for chunk in vals.chunks(COMPRESSION_BLOCK_SIZE) {
let compressed_block = encoder.compress_block_unsorted(chunk);
let num_bits = compressed_block[0];
assert_eq!(compressed_block_size(num_bits), compressed_block.len());
buffer.extend_from_slice(compressed_block);
}
if cfg!(simd) {
buffer.extend_from_slice(&[0u8; 7]);
}
ReadOnlySource::from(buffer)
}
#[test]
fn test_compressed_int_stream() {
let buffer = create_stream_buffer();
let mut stream = CompressedIntStream::wrap(buffer);
let mut block: [u32; COMPRESSION_BLOCK_SIZE] = [0u32; COMPRESSION_BLOCK_SIZE];
stream.read(&mut block[0..2]);
assert_eq!(block[0], 0);
assert_eq!(block[1], 1);
stream.skip(5);
stream.read(&mut block[0..3]);
assert_eq!(block[0], 7);
assert_eq!(block[1], 8);
assert_eq!(block[2], 9);
stream.skip(500);
stream.read(&mut block[0..3]);
assert_eq!(block[0], 510);
assert_eq!(block[1], 511);
assert_eq!(block[2], 512);
stream.skip(511);
stream.read(&mut block[..1]);
assert_eq!(block[0], 1024);
}
}

92
src/compression/vint/compression_vint_nosimd.rs Normal file
View File

@@ -0,0 +1,92 @@
#[inline(always)]
pub(crate) fn compress_sorted<'a>(
input: &[u32],
output: &'a mut [u8],
mut offset: u32,
) -> &'a [u8] {
let mut byte_written = 0;
for &v in input {
let mut to_encode: u32 = v - offset;
offset = v;
loop {
let next_byte: u8 = (to_encode % 128u32) as u8;
to_encode /= 128u32;
if to_encode == 0u32 {
output[byte_written] = next_byte | 128u8;
byte_written += 1;
break;
} else {
output[byte_written] = next_byte;
byte_written += 1;
}
}
}
&output[..byte_written]
}
#[inline(always)]
pub(crate) fn compress_unsorted<'a>(input: &[u32], output: &'a mut [u8]) -> &'a [u8] {
let mut byte_written = 0;
for &v in input {
let mut to_encode: u32 = v;
loop {
let next_byte: u8 = (to_encode % 128u32) as u8;
to_encode /= 128u32;
if to_encode == 0u32 {
output[byte_written] = next_byte | 128u8;
byte_written += 1;
break;
} else {
output[byte_written] = next_byte;
byte_written += 1;
}
}
}
&output[..byte_written]
}
#[inline(always)]
pub(crate) fn uncompress_sorted<'a>(
compressed_data: &'a [u8],
output: &mut [u32],
offset: u32,
) -> usize {
let mut read_byte = 0;
let mut result = offset;
let num_els = output.len();
for i in 0..num_els {
let mut shift = 0u32;
loop {
let cur_byte = compressed_data[read_byte];
read_byte += 1;
result += ((cur_byte % 128u8) as u32) << shift;
if cur_byte & 128u8 != 0u8 {
break;
}
shift += 7;
}
output[i] = result;
}
read_byte
}
#[inline(always)]
pub(crate) fn uncompress_unsorted<'a>(compressed_data: &'a [u8], output: &mut [u32]) -> usize {
let mut read_byte = 0;
let num_els = output.len();
for i in 0..num_els {
let mut result = 0u32;
let mut shift = 0u32;
loop {
let cur_byte = compressed_data[read_byte];
read_byte += 1;
result += ((cur_byte % 128u8) as u32) << shift;
if cur_byte & 128u8 != 0u8 {
break;
}
shift += 7;
}
output[i] = result;
}
read_byte
}

72
src/compression/vint/compression_vint_simd.rs Normal file
View File

@@ -0,0 +1,72 @@
mod streamvbyte {
use libc::size_t;
extern "C" {
pub fn streamvbyte_delta_encode(
data: *const u32,
num_els: u32,
output: *mut u8,
offset: u32,
) -> size_t;
pub fn streamvbyte_delta_decode(
compressed_data: *const u8,
output: *mut u32,
num_els: u32,
offset: u32,
) -> size_t;
pub fn streamvbyte_encode(data: *const u32, num_els: u32, output: *mut u8) -> size_t;
pub fn streamvbyte_decode(
compressed_data: *const u8,
output: *mut u32,
num_els: usize,
) -> size_t;
}
}
#[inline(always)]
pub(crate) fn compress_sorted<'a>(input: &[u32], output: &'a mut [u8], offset: u32) -> &'a [u8] {
let compress_length = unsafe {
streamvbyte::streamvbyte_delta_encode(
input.as_ptr(),
input.len() as u32,
output.as_mut_ptr(),
offset,
)
};
&output[..compress_length]
}
#[inline(always)]
pub(crate) fn compress_unsorted<'a>(input: &[u32], output: &'a mut [u8]) -> &'a [u8] {
let compress_length = unsafe {
streamvbyte::streamvbyte_encode(input.as_ptr(), input.len() as u32, output.as_mut_ptr())
};
&output[..compress_length]
}
#[inline(always)]
pub(crate) fn uncompress_sorted<'a>(
compressed_data: &'a [u8],
output: &mut [u32],
offset: u32,
) -> usize {
unsafe {
streamvbyte::streamvbyte_delta_decode(
compressed_data.as_ptr(),
output.as_mut_ptr(),
output.len() as u32,
offset,
)
}
}
#[inline(always)]
pub(crate) fn uncompress_unsorted<'a>(compressed_data: &'a [u8], output: &mut [u32]) -> usize {
unsafe {
streamvbyte::streamvbyte_decode(compressed_data.as_ptr(), output.as_mut_ptr(), output.len())
}
}

224
src/core/index.rs
View File

@@ -1,51 +1,45 @@
use Result;
use Error;
use error::{ErrorKind, ResultExt};
use serde_json;
use schema::Schema;
use std::sync::Arc;
use std::borrow::BorrowMut;
use std::fmt;
use rustc_serialize::json;
use core::SegmentId;
use directory::{Directory, MmapDirectory, RAMDirectory};
use indexer::IndexWriter;
use indexer::index_writer::open_index_writer;
use core::searcher::Searcher;
use std::convert::From;
use num_cpus;
use super::segment::Segment;
use core::SegmentReader;
use super::pool::Pool;
use core::SegmentMeta;
use super::pool::LeasedItem;
use std::path::Path;
use indexer::SegmentManager;
use core::IndexMeta;
use indexer::DirectoryLock;
use IndexWriter;
use directory::ManagedDirectory;
use core::META_FILEPATH;
use super::segment::create_segment;
use indexer::segment_updater::save_new_metas;
use tokenizer::TokenizerManager;
const NUM_SEARCHERS: usize = 12;
/// Accessor to the index segment manager
///
/// This method is not part of tantivy's public API
pub fn get_segment_manager(index: &Index) -> Arc<SegmentManager> {
index.segment_manager.clone()
}
fn load_metas(directory: &Directory) -> Result<IndexMeta> {
let meta_file = try!(directory.open_read(&META_FILEPATH));
let meta_content = String::from_utf8_lossy(meta_file.as_slice());
json::decode(&meta_content)
.map_err(|e| Error::CorruptedFile(META_FILEPATH.clone(), Box::new(e)))
let meta_data = directory.atomic_read(&META_FILEPATH)?;
let meta_string = String::from_utf8_lossy(&meta_data);
serde_json::from_str(&meta_string).chain_err(|| ErrorKind::CorruptedFile(META_FILEPATH.clone()))
}
/// Tantivy's Search Index
/// Search Index
pub struct Index {
segment_manager: Arc<SegmentManager>,
directory: Box<Directory>,
directory: ManagedDirectory,
schema: Schema,
searcher_pool: Arc<Pool<Searcher>>,
docstamp: u64,
tokenizers: TokenizerManager,
}
impl Index {
@@ -54,18 +48,28 @@ impl Index {
/// The index will be allocated in anonymous memory.
/// This should only be used for unit tests.
pub fn create_in_ram(schema: Schema) -> Index {
let directory = Box::new(RAMDirectory::create());
Index::from_directory(directory, schema).expect("Creating a RAMDirectory should never fail") // unwrap is ok here
let ram_directory = RAMDirectory::create();
// unwrap is ok here
let directory = ManagedDirectory::new(ram_directory).expect(
"Creating a managed directory from a brand new RAM directory \
should never fail.",
);
Index::from_directory(directory, schema).expect("Creating a RAMDirectory should never fail")
}
/// Creates a new index in a given filepath.
/// The index will use the `MMapDirectory`.
///
/// If a previous index was in this directory, then its meta file will be destroyed.
pub fn create(directory_path: &Path, schema: Schema) -> Result<Index> {
let mut directory = MmapDirectory::open(directory_path)?;
save_new_metas(schema.clone(), 0, &mut directory)?;
Index::from_directory(box directory, schema)
pub fn create<P: AsRef<Path>>(directory_path: P, schema: Schema) -> Result<Index> {
let mmap_directory = MmapDirectory::open(directory_path)?;
let directory = ManagedDirectory::new(mmap_directory)?;
Index::from_directory(directory, schema)
}
/// Accessor for the tokenizer manager.
pub fn tokenizers(&self) -> &TokenizerManager {
&self.tokenizers
}
/// Creates a new index in a temp directory.
@@ -77,64 +81,71 @@ impl Index {
/// The temp directory is only used for testing the `MmapDirectory`.
/// For other unit tests, prefer the `RAMDirectory`, see: `create_in_ram`.
pub fn create_from_tempdir(schema: Schema) -> Result<Index> {
let directory = Box::new(try!(MmapDirectory::create_from_tempdir()));
let mmap_directory = MmapDirectory::create_from_tempdir()?;
let directory = ManagedDirectory::new(mmap_directory)?;
Index::from_directory(directory, schema)
}
/// Creates a new index given a directory and an `IndexMeta`.
fn create_from_metas(directory: Box<Directory>, metas: IndexMeta) -> Result<Index> {
fn create_from_metas(directory: ManagedDirectory, metas: &IndexMeta) -> Result<Index> {
let schema = metas.schema.clone();
let docstamp = metas.docstamp;
let committed_segments = metas.committed_segments;
// TODO log somethings is uncommitted is not empty.
let index = Index {
segment_manager: Arc::new(SegmentManager::from_segments(committed_segments)),
directory: directory,
schema: schema,
directory,
schema,
searcher_pool: Arc::new(Pool::new()),
docstamp: docstamp,
tokenizers: TokenizerManager::default(),
};
try!(index.load_searchers());
index.load_searchers()?;
Ok(index)
}
/// Opens a new directory from a directory.
pub fn from_directory(directory: Box<Directory>, schema: Schema) -> Result<Index> {
Index::create_from_metas(directory, IndexMeta::with_schema(schema))
/// Create a new index from a directory.
pub fn from_directory(mut directory: ManagedDirectory, schema: Schema) -> Result<Index> {
save_new_metas(schema.clone(), 0, directory.borrow_mut())?;
let metas = IndexMeta::with_schema(schema);
Index::create_from_metas(directory, &metas)
}
/// Opens a new directory from an index path.
pub fn open(directory_path: &Path) -> Result<Index> {
let directory = try!(MmapDirectory::open(directory_path));
let metas = try!(load_metas(&directory)); //< TODO does the directory already exists?
Index::create_from_metas(directory.box_clone(), metas)
pub fn open<P: AsRef<Path>>(directory_path: P) -> Result<Index> {
let mmap_directory = MmapDirectory::open(directory_path)?;
let directory = ManagedDirectory::new(mmap_directory)?;
let metas = load_metas(&directory)?;
Index::create_from_metas(directory, &metas)
}
/// Returns the index docstamp.
/// Reads the index meta file from the directory.
pub fn load_metas(&self) -> Result<IndexMeta> {
load_metas(self.directory())
}
/// Open a new index writer. Attempts to acquire a lockfile.
///
/// The docstamp is the number of documents that have been added
/// from the beginning of time, and until the moment of the last commit.
pub fn docstamp(&self) -> u64 {
self.docstamp
}
/// Creates a multithreaded writer.
/// Each writer produces an independent segment.
/// The lockfile should be deleted on drop, but it is possible
/// that due to a panic or other error, a stale lockfile will be
/// left in the index directory. If you are sure that no other
/// `IndexWriter` on the system is accessing the index directory,
/// it is safe to manually delete the lockfile.
///
/// num_threads specifies the number of indexing workers that
/// should work at the same time.
///
/// # Errors
/// If the lockfile already exists, returns `Error::FileAlreadyExists`.
/// # Panics
/// If the heap size per thread is too small, panics.
pub fn writer_with_num_threads(&self,
num_threads: usize,
heap_size_in_bytes: usize)
-> Result<IndexWriter> {
IndexWriter::open(self, num_threads, heap_size_in_bytes)
pub fn writer_with_num_threads(
&self,
num_threads: usize,
heap_size_in_bytes: usize,
) -> Result<IndexWriter> {
let directory_lock = DirectoryLock::lock(self.directory().box_clone())?;
open_index_writer(self, num_threads, heap_size_in_bytes, directory_lock)
}
/// Creates a multithreaded writer
/// It just calls `writer_with_num_threads` with the number of cores as `num_threads`
///
/// # Errors
/// If the lockfile already exists, returns `Error::FileAlreadyExists`.
/// # Panics
@@ -151,65 +162,63 @@ impl Index {
}
/// Returns the list of segments that are searchable
pub fn searchable_segments(&self) -> Vec<Segment> {
self.searchable_segment_ids()
pub fn searchable_segments(&self) -> Result<Vec<Segment>> {
Ok(self.searchable_segment_metas()?
.into_iter()
.map(|segment_id| self.segment(segment_id))
.collect()
.map(|segment_meta| self.segment(segment_meta))
.collect())
}
/// Remove all of the file associated with the segment.
///
/// This method cannot fail. If a problem occurs,
/// some files may end up never being removed.
/// The error will only be logged.
pub fn delete_segment(&self, segment_id: SegmentId) {
self.segment(segment_id).delete();
}
/// Return a segment object given a `segment_id`
///
/// The segment may or may not exist.
pub fn segment(&self, segment_id: SegmentId) -> Segment {
create_segment(self.clone(), segment_id)
}
/// Return a reference to the index directory.
pub fn directory(&self) -> &Directory {
&*self.directory
}
/// Return a mutable reference to the index directory.
pub fn directory_mut(&mut self) -> &mut Directory {
&mut *self.directory
}
/// Returns the list of segment ids that are searchable.
fn searchable_segment_ids(&self) -> Vec<SegmentId> {
self.segment_manager.committed_segments()
#[doc(hidden)]
pub fn segment(&self, segment_meta: SegmentMeta) -> Segment {
create_segment(self.clone(), segment_meta)
}
/// Creates a new segment.
pub fn new_segment(&self) -> Segment {
self.segment(SegmentId::generate_random())
let segment_meta = SegmentMeta::new(SegmentId::generate_random());
create_segment(self.clone(), segment_meta)
}
/// Return a reference to the index directory.
pub fn directory(&self) -> &ManagedDirectory {
&self.directory
}
/// Return a mutable reference to the index directory.
pub fn directory_mut(&mut self) -> &mut ManagedDirectory {
&mut self.directory
}
/// Reads the meta.json and returns the list of
/// `SegmentMeta` from the last commit.
pub fn searchable_segment_metas(&self) -> Result<Vec<SegmentMeta>> {
Ok(self.load_metas()?.segments)
}
/// Returns the list of segment ids that are searchable.
pub fn searchable_segment_ids(&self) -> Result<Vec<SegmentId>> {
Ok(self.searchable_segment_metas()?
.iter()
.map(|segment_meta| segment_meta.id())
.collect())
}
/// Creates a new generation of searchers after
/// a change of the set of searchable indexes.
///
/// This needs to be called when a new segment has been
/// published or after a merge.
pub fn load_searchers(&self) -> Result<()> {
let searchable_segments = self.searchable_segments();
let mut searchers = Vec::new();
for _ in 0..NUM_SEARCHERS {
let searchable_segments_clone = searchable_segments.clone();
let segment_readers: Vec<SegmentReader> = try!(searchable_segments_clone.into_iter()
.map(SegmentReader::open)
.collect());
let searcher = Searcher::from(segment_readers);
searchers.push(searcher);
}
let searchable_segments = self.searchable_segments()?;
let segment_readers: Vec<SegmentReader> = searchable_segments
.iter()
.map(SegmentReader::open)
.collect::<Result<_>>()?;
let searchers = (0..NUM_SEARCHERS)
.map(|_| Searcher::from(segment_readers.clone()))
.collect();
self.searcher_pool.publish_new_generation(searchers);
Ok(())
}
@@ -229,7 +238,6 @@ impl Index {
}
}
impl fmt::Debug for Index {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(f, "Index({:?})", self.directory)
@@ -239,12 +247,10 @@ impl fmt::Debug for Index {
impl Clone for Index {
fn clone(&self) -> Index {
Index {
segment_manager: self.segment_manager.clone(),
directory: self.directory.box_clone(),
directory: self.directory.clone(),
schema: self.schema.clone(),
searcher_pool: self.searcher_pool.clone(),
docstamp: self.docstamp,
searcher_pool: Arc::clone(&self.searcher_pool),
tokenizers: self.tokenizers.clone(),
}
}
}

61
src/core/index_meta.rs
View File

@@ -1,47 +1,54 @@
use schema::Schema;
use core::SegmentId;
use core::SegmentMeta;
/// Meta information about the `Index`.
///
///
/// This object is serialized on disk in the `meta.json` file.
/// It keeps information about
/// It keeps information about
/// * the searchable segments,
/// * the index docstamp
/// * the index `docstamp`
/// * the schema
///
#[derive(Clone,Debug,RustcDecodable,RustcEncodable)]
#[derive(Clone, Debug, Serialize, Deserialize)]
pub struct IndexMeta {
pub committed_segments: Vec<SegmentMeta>,
pub uncommitted_segments: Vec<SegmentMeta>,
pub segments: Vec<SegmentMeta>,
pub schema: Schema,
pub docstamp: u64,
pub opstamp: u64,
#[serde(skip_serializing_if = "Option::is_none")] pub payload: Option<String>,
}
impl IndexMeta {
pub fn with_schema(schema: Schema) -> IndexMeta {
IndexMeta {
committed_segments: Vec::new(),
uncommitted_segments: Vec::new(),
schema: schema,
docstamp: 0u64,
segments: vec![],
schema,
opstamp: 0u64,
payload: None,
}
}
}
#[derive(Clone, Debug, RustcDecodable,RustcEncodable)]
pub struct SegmentMeta {
pub segment_id: SegmentId,
pub num_docs: u32,
}
#[cfg(test)]
impl SegmentMeta {
pub fn new(segment_id: SegmentId, num_docs: u32) -> SegmentMeta {
SegmentMeta {
segment_id: segment_id,
num_docs: num_docs,
}
mod tests {
use serde_json;
use super::IndexMeta;
use schema::{SchemaBuilder, TEXT};
#[test]
fn test_serialize_metas() {
let schema = {
let mut schema_builder = SchemaBuilder::new();
schema_builder.add_text_field("text", TEXT);
schema_builder.build()
};
let index_metas = IndexMeta {
segments: Vec::new(),
schema: schema,
opstamp: 0u64,
payload: None,
};
let json = serde_json::ser::to_string(&index_metas).expect("serialization failed");
assert_eq!(json, r#"{"segments":[],"schema":[{"name":"text","type":"text","options":{"indexing":{"record":"position","tokenizer":"default"},"stored":false}}],"opstamp":0}"#);
}
}
}

152
src/core/inverted_index_reader.rs Normal file
View File

@@ -0,0 +1,152 @@
use directory::{ReadOnlySource, SourceRead};
use termdict::{TermDictionary, TermDictionaryImpl};
use postings::{BlockSegmentPostings, SegmentPostings};
use postings::TermInfo;
use schema::IndexRecordOption;
use schema::Term;
use std::cmp;
use fastfield::DeleteBitSet;
use schema::Schema;
use compression::CompressedIntStream;
/// The inverted index reader is in charge of accessing
/// the inverted index associated to a specific field.
///
/// # Note
///
/// It is safe to delete the segment associated to
/// an `InvertedIndexReader`. As long as it is open,
/// the `ReadOnlySource` it is relying on should
/// stay available.
///
///
/// `InvertedIndexReader` are created by calling
/// the `SegmentReader`'s [`.inverted_index(...)`] method
pub struct InvertedIndexReader {
termdict: TermDictionaryImpl,
postings_source: ReadOnlySource,
positions_source: ReadOnlySource,
delete_bitset: DeleteBitSet,
schema: Schema,
}
impl InvertedIndexReader {
pub(crate) fn new(
termdict_source: ReadOnlySource,
postings_source: ReadOnlySource,
positions_source: ReadOnlySource,
delete_bitset: DeleteBitSet,
schema: Schema,
) -> InvertedIndexReader {
InvertedIndexReader {
termdict: TermDictionaryImpl::from_source(termdict_source),
postings_source,
positions_source,
delete_bitset,
schema,
}
}
/// Returns the term info associated with the term.
pub fn get_term_info(&self, term: &Term) -> Option<TermInfo> {
self.termdict.get(term.as_slice())
}
/// Return the term dictionary datastructure.
pub fn terms(&self) -> &TermDictionaryImpl {
&self.termdict
}
/// Resets the block segment to another position of the postings
/// file.
///
/// This is useful for enumerating through a list of terms,
/// and consuming the associated posting lists while avoiding
/// reallocating a `BlockSegmentPostings`.
///
/// # Warning
///
/// This does not reset the positions list.
pub fn reset_block_postings_from_terminfo(
&self,
term_info: &TermInfo,
block_postings: &mut BlockSegmentPostings,
) {
let offset = term_info.postings_offset as usize;
let end_source = self.postings_source.len();
let postings_slice = self.postings_source.slice(offset, end_source);
let postings_reader = SourceRead::from(postings_slice);
block_postings.reset(term_info.doc_freq as usize, postings_reader);
}
/// Returns a block postings given a `term_info`.
/// This method is for an advanced usage only.
///
/// Most user should prefer using `read_postings` instead.
pub fn read_block_postings_from_terminfo(
&self,
term_info: &TermInfo,
option: IndexRecordOption,
) -> BlockSegmentPostings {
let offset = term_info.postings_offset as usize;
let postings_data = self.postings_source.slice_from(offset);
let has_freq = option.has_freq();
BlockSegmentPostings::from_data(
term_info.doc_freq as usize,
SourceRead::from(postings_data),
has_freq,
)
}
/// Returns a posting object given a `term_info`.
/// This method is for an advanced usage only.
///
/// Most user should prefer using `read_postings` instead.
pub fn read_postings_from_terminfo(
&self,
term_info: &TermInfo,
option: IndexRecordOption,
) -> SegmentPostings {
let block_postings = self.read_block_postings_from_terminfo(term_info, option);
let delete_bitset = self.delete_bitset.clone();
let position_stream = {
if option.has_positions() {
let position_offset = term_info.positions_offset;
let positions_source = self.positions_source.slice_from(position_offset as usize);
let mut stream = CompressedIntStream::wrap(positions_source);
stream.skip(term_info.positions_inner_offset as usize);
Some(stream)
} else {
None
}
};
SegmentPostings::from_block_postings(block_postings, delete_bitset, position_stream)
}
/// Returns the segment postings associated with the term, and with the given option,
/// or `None` if the term has never been encountered and indexed.
///
/// If the field was not indexed with the indexing options that cover
/// the requested options, the returned `SegmentPostings` the method does not fail
/// and returns a `SegmentPostings` with as much information as possible.
///
/// For instance, requesting `IndexRecordOption::Freq` for a
/// `TextIndexingOptions` that does not index position will return a `SegmentPostings`
/// with `DocId`s and frequencies.
pub fn read_postings(&self, term: &Term, option: IndexRecordOption) -> Option<SegmentPostings> {
let field = term.field();
let field_entry = self.schema.get_field_entry(field);
let term_info = get!(self.get_term_info(term));
let maximum_option = get!(field_entry.field_type().get_index_record_option());
let best_effort_option = cmp::min(maximum_option, option);
Some(self.read_postings_from_terminfo(&term_info, best_effort_option))
}
/// Returns the number of documents containing the term.
pub fn doc_freq(&self, term: &Term) -> u32 {
match self.get_term_info(term) {
Some(term_info) => term_info.doc_freq,
None => 0,
}
}
}

30
src/core/mod.rs
View File

@@ -1,5 +1,4 @@
pub mod searcher;
pub mod index;
mod segment_reader;
mod segment_id;
@@ -7,20 +6,37 @@ mod segment_component;
mod segment;
mod index_meta;
mod pool;
mod segment_meta;
mod inverted_index_reader;
use std::path::PathBuf;
pub use self::inverted_index_reader::InvertedIndexReader;
pub use self::searcher::Searcher;
pub use self::segment_component::SegmentComponent;
pub use self::segment_id::SegmentId;
pub use self::segment_reader::SegmentReader;
pub use self::segment::Segment;
pub use self::segment::SegmentInfo;
pub use self::segment::SerializableSegment;
pub use self::index::Index;
pub use self::index_meta::{IndexMeta, SegmentMeta};
pub use self::segment_meta::SegmentMeta;
pub use self::index_meta::IndexMeta;
use std::path::PathBuf;
lazy_static! {
/// The meta file contains all the information about the list of segments and the schema
/// of the index.
pub static ref META_FILEPATH: PathBuf = PathBuf::from("meta.json");
}
/// The managed file contains a list of files that were created by the tantivy
/// and will therefore be garbage collected when they are deemed useless by tantivy.
///
/// Removing this file is safe, but will prevent the garbage collection of all of the file that
/// are currently in the directory
pub static ref MANAGED_FILEPATH: PathBuf = PathBuf::from(".managed.json");
/// Only one process should be able to write tantivy's index at a time.
/// This file, when present, is in charge of preventing other processes to open an IndexWriter.
///
/// If the process is killed and this file remains, it is safe to remove it manually.
pub static ref LOCKFILE_FILEPATH: PathBuf = PathBuf::from(".tantivy-indexer.lock");
}

56
src/core/pool.rs
View File

@@ -17,10 +17,10 @@ pub struct Pool<T> {
}
impl<T> Pool<T> {
pub fn new() -> Pool<T> {
let queue = Arc::new(MsQueue::new());
Pool {
queue: Arc::new(MsQueue::new()),
queue,
freshest_generation: AtomicUsize::default(),
next_generation: AtomicUsize::default(),
}
@@ -30,52 +30,52 @@ impl<T> Pool<T> {
let next_generation = self.next_generation.fetch_add(1, Ordering::SeqCst) + 1;
for item in items {
let gen_item = GenerationItem {
item: item,
item,
generation: next_generation,
};
self.queue.push(gen_item);
}
self.advertise_generation(next_generation);
}
/// At the exit of this method,
/// At the exit of this method,
/// - freshest_generation has a value greater or equal than generation
/// - freshest_generation has a value that has been advertised
/// - freshest_generation has
/// - freshest_generation has)
fn advertise_generation(&self, generation: usize) {
// not optimal at all but the easiest to read proof.
// not optimal at all but the easiest to read proof.
loop {
let former_generation = self.freshest_generation.load(Ordering::Acquire);
if former_generation >= generation {
break;
}
self.freshest_generation.compare_and_swap(former_generation, generation, Ordering::SeqCst);
}
self.freshest_generation.compare_and_swap(
former_generation,
generation,
Ordering::SeqCst,
);
}
}
fn generation(&self,) -> usize {
fn generation(&self) -> usize {
self.freshest_generation.load(Ordering::Acquire)
}
pub fn acquire(&self,) -> LeasedItem<T> {
pub fn acquire(&self) -> LeasedItem<T> {
let generation = self.generation();
loop {
let gen_item = self.queue.pop();
if gen_item.generation >= generation {
return LeasedItem {
gen_item: Some(gen_item),
recycle_queue: self.queue.clone(),
}
}
else {
recycle_queue: Arc::clone(&self.queue),
};
} else {
// this searcher is obsolete,
// removing it from the pool.
}
}
}
}
pub struct LeasedItem<T> {
@@ -84,29 +84,33 @@ pub struct LeasedItem<T> {
}
impl<T> Deref for LeasedItem<T> {
type Target = T;
fn deref(&self) -> &T {
&self.gen_item.as_ref().expect("Unwrapping a leased item should never fail").item // unwrap is safe here
&self.gen_item
.as_ref()
.expect("Unwrapping a leased item should never fail")
.item // unwrap is safe here
}
}
impl<T> DerefMut for LeasedItem<T> {
fn deref_mut(&mut self) -> &mut T {
&mut self.gen_item.as_mut().expect("Unwrapping a mut leased item should never fail").item // unwrap is safe here
&mut self.gen_item
.as_mut()
.expect("Unwrapping a mut leased item should never fail")
.item // unwrap is safe here
}
}
impl<T> Drop for LeasedItem<T> {
fn drop(&mut self) {
let gen_item: GenerationItem<T> = mem::replace(&mut self.gen_item, None).expect("Unwrapping a leased item should never fail");
let gen_item: GenerationItem<T> = mem::replace(&mut self.gen_item, None)
.expect("Unwrapping a leased item should never fail");
self.recycle_queue.push(gen_item);
}
}
#[cfg(test)]
mod tests {
@@ -127,4 +131,4 @@ mod tests {
assert_eq!(*pool.acquire(), 11);
}
}
}
}

82
src/core/searcher.rs
View File

@@ -6,68 +6,106 @@ use common::TimerTree;
use query::Query;
use DocId;
use DocAddress;
use schema::Term;
use schema::{Field, Term};
use termdict::{TermDictionary, TermMerger};
use std::sync::Arc;
use std::fmt;
use core::InvertedIndexReader;
/// Holds a list of `SegmentReader`s ready for search.
///
/// It guarantees that the `Segment` will not be removed before
/// It guarantees that the `Segment` will not be removed before
/// the destruction of the `Searcher`.
///
#[derive(Debug)]
///
pub struct Searcher {
segment_readers: Vec<SegmentReader>,
}
impl Searcher {
/// Fetches a document from tantivy's store given a `DocAddress`.
///
/// The searcher uses the segment ordinal to route the
/// the request to the right `Segment`.
/// the request to the right `Segment`.
pub fn doc(&self, doc_address: &DocAddress) -> Result<Document> {
let DocAddress(segment_local_id, doc_id) = *doc_address;
let segment_reader = &self.segment_readers[segment_local_id as usize];
segment_reader.doc(doc_id)
}
/// Returns the overall number of documents in the index.
pub fn num_docs(&self,) -> DocId {
pub fn num_docs(&self) -> DocId {
self.segment_readers
.iter()
.map(|segment_reader| segment_reader.num_docs())
.fold(0u32, |acc, val| acc + val)
}
/// Return the overall number of documents containing
/// the given term.
/// the given term.
pub fn doc_freq(&self, term: &Term) -> u32 {
self.segment_readers
.iter()
.map(|segment_reader| segment_reader.doc_freq(term))
.map(|segment_reader| segment_reader.inverted_index(term.field()).doc_freq(term))
.fold(0u32, |acc, val| acc + val)
}
/// Return the list of segment readers
pub fn segment_readers(&self,) -> &Vec<SegmentReader> {
pub fn segment_readers(&self) -> &[SegmentReader] {
&self.segment_readers
}
/// Returns the segment_reader associated with the given segment_ordinal
pub fn segment_reader(&self, segment_ord: usize) -> &SegmentReader {
&self.segment_readers[segment_ord]
pub fn segment_reader(&self, segment_ord: u32) -> &SegmentReader {
&self.segment_readers[segment_ord as usize]
}
/// Runs a query on the segment readers wrapped by the searcher
pub fn search<C: Collector>(&self, query: &Query, collector: &mut C) -> Result<TimerTree> {
query.search(self, collector)
}
/// Return the field searcher associated to a `Field`.
pub fn field(&self, field: Field) -> FieldSearcher {
let inv_index_readers = self.segment_readers
.iter()
.map(|segment_reader| segment_reader.inverted_index(field))
.collect::<Vec<_>>();
FieldSearcher::new(inv_index_readers)
}
}
pub struct FieldSearcher {
inv_index_readers: Vec<Arc<InvertedIndexReader>>,
}
impl FieldSearcher {
fn new(inv_index_readers: Vec<Arc<InvertedIndexReader>>) -> FieldSearcher {
FieldSearcher { inv_index_readers }
}
/// Returns a Stream over all of the sorted unique terms of
/// for the given field.
pub fn terms(&self) -> TermMerger {
let term_streamers: Vec<_> = self.inv_index_readers
.iter()
.map(|inverted_index| inverted_index.terms().stream())
.collect();
TermMerger::new(term_streamers)
}
}
impl From<Vec<SegmentReader>> for Searcher {
fn from(segment_readers: Vec<SegmentReader>) -> Searcher {
Searcher {
segment_readers: segment_readers,
}
Searcher { segment_readers }
}
}
}
impl fmt::Debug for Searcher {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
let segment_ids = self.segment_readers
.iter()
.map(|segment_reader| segment_reader.segment_id())
.collect::<Vec<_>>();
write!(f, "Searcher({:?})", segment_ids)
}
}

137
src/core/segment.rs
View File

@@ -1,100 +1,98 @@
use Result;
use std::path::PathBuf;
use schema::Schema;
use DocId;
use std::fmt;
use core::SegmentId;
use directory::{ReadOnlySource, WritePtr};
use directory::{FileProtection, ReadOnlySource, WritePtr};
use indexer::segment_serializer::SegmentSerializer;
use super::SegmentComponent;
use core::Index;
use std::result;
use directory::error::{FileError, OpenWriteError};
use directory::Directory;
use core::SegmentMeta;
use directory::error::{OpenReadError, OpenWriteError};
/// A segment is a piece of the index.
#[derive(Clone)]
pub struct Segment {
index: Index,
segment_id: SegmentId,
meta: SegmentMeta,
}
impl fmt::Debug for Segment {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(f, "Segment({:?})", self.segment_id.uuid_string())
write!(f, "Segment({:?})", self.id().uuid_string())
}
}
/// Creates a new segment given an `Index` and a `SegmentId`
///
/// The function is here to make it private outside `tantivy`.
pub fn create_segment(index: Index, segment_id: SegmentId) -> Segment {
Segment {
index: index,
segment_id: segment_id,
}
///
/// The function is here to make it private outside `tantivy`.
pub fn create_segment(index: Index, meta: SegmentMeta) -> Segment {
Segment { index, meta }
}
impl Segment {
/// Returns the index the segment belongs to.
pub fn index(&self) -> &Index {
&self.index
}
/// Returns our index's schema.
pub fn schema(&self,) -> Schema {
pub fn schema(&self) -> Schema {
self.index.schema()
}
/// Returns the segment's id.
pub fn id(&self,) -> SegmentId {
self.segment_id
/// Returns the segment meta-information
pub fn meta(&self) -> &SegmentMeta {
&self.meta
}
#[doc(hidden)]
pub fn set_delete_meta(&mut self, num_deleted_docs: u32, opstamp: u64) {
self.meta.set_delete_meta(num_deleted_docs, opstamp);
}
/// Returns the segment's id.
pub fn id(&self) -> SegmentId {
self.meta.id()
}
/// Returns the relative path of a component of our segment.
///
/// It just joins the segment id with the extension
///
/// It just joins the segment id with the extension
/// associated to a segment component.
pub fn relative_path(&self, component: SegmentComponent) -> PathBuf {
self.segment_id.relative_path(component)
self.meta.relative_path(component)
}
/// Deletes all of the document of the segment.
/// This is called when there is a merge or a rollback.
/// Protects a specific component file from being deleted.
///
/// # Disclaimer
/// If deletion of a file fails (e.g. a file
/// was read-only.), the method does not
/// fail and just logs an error
pub fn delete(&self,) {
for component in SegmentComponent::values() {
let rel_path = self.relative_path(component);
if let Err(err) = self.index.directory().delete(&rel_path) {
match err {
FileError::FileDoesNotExist(_) => {
// this is normal behavior.
// the position file for instance may not exists.
}
FileError::IOError(err) => {
error!("Failed to remove {:?} : {:?}", self.segment_id, err);
}
}
}
}
/// Returns a FileProtection object. The file is guaranteed
/// to not be garbage collected as long as this `FileProtection` object
/// lives.
pub fn protect_from_delete(&self, component: SegmentComponent) -> FileProtection {
let path = self.relative_path(component);
self.index.directory().protect_file_from_delete(&path)
}
/// Open one of the component file for read.
pub fn open_read(&self, component: SegmentComponent) -> result::Result<ReadOnlySource, FileError> {
/// Open one of the component file for a *regular* read.
pub fn open_read(
&self,
component: SegmentComponent,
) -> result::Result<ReadOnlySource, OpenReadError> {
let path = self.relative_path(component);
let source = try!(self.index.directory().open_read(&path));
let source = self.index.directory().open_read(&path)?;
Ok(source)
}
/// Open one of the component file for write.
pub fn open_write(&mut self, component: SegmentComponent) -> result::Result<WritePtr, OpenWriteError> {
/// Open one of the component file for *regular* write.
pub fn open_write(
&mut self,
component: SegmentComponent,
) -> result::Result<WritePtr, OpenWriteError> {
let path = self.relative_path(component);
let write = try!(self.index.directory_mut().open_write(&path));
let write = self.index.directory_mut().open_write(&path)?;
Ok(write)
}
}
@@ -108,7 +106,34 @@ pub trait SerializableSegment {
fn write(&self, serializer: SegmentSerializer) -> Result<u32>;
}
#[derive(Clone,Debug,RustcDecodable,RustcEncodable)]
pub struct SegmentInfo {
pub max_doc: DocId,
}
#[cfg(test)]
mod tests {
use core::SegmentComponent;
use directory::Directory;
use std::collections::HashSet;
use schema::SchemaBuilder;
use Index;
#[test]
fn test_segment_protect_component() {
let mut index = Index::create_in_ram(SchemaBuilder::new().build());
let segment = index.new_segment();
let path = segment.relative_path(SegmentComponent::POSTINGS);
let directory = index.directory_mut();
directory.atomic_write(&*path, &vec![0u8]).unwrap();
let living_files = HashSet::new();
{
let _file_protection = segment.protect_from_delete(SegmentComponent::POSTINGS);
assert!(directory.exists(&*path));
directory.garbage_collect(|| living_files.clone());
assert!(directory.exists(&*path));
}
directory.garbage_collect(|| living_files);
assert!(!directory.exists(&*path));
}
}

44
src/core/segment_component.rs
View File

@@ -1,41 +1,41 @@
use std::vec::IntoIter;
/// Enum describing each component of a tantivy segment.
/// Each component is stored in its own file,
/// using the pattern `segment_uuid`.`component_extension`,
/// except the delete component that takes an `segment_uuid`.`delete_opstamp`.`component_extension`
#[derive(Copy, Clone)]
pub enum SegmentComponent {
INFO,
/// Postings (or inverted list). Sorted lists of document ids, associated to terms
POSTINGS,
/// Positions of terms in each document.
POSITIONS,
/// Column-oriented random-access storage of fields.
FASTFIELDS,
/// Stores the sum of the length (in terms) of each field for each document.
/// Field norms are stored as a special u64 fast field.
FIELDNORMS,
/// Dictionary associating `Term`s to `TermInfo`s which is
/// simply an address into the `postings` file and the `positions` file.
TERMS,
/// Row-oriented, LZ4-compressed storage of the documents.
/// Accessing a document from the store is relatively slow, as it
/// requires to decompress the entire block it belongs to.
STORE,
/// Bitset describing which document of the segment is deleted.
DELETE,
}
impl SegmentComponent {
pub fn values() -> IntoIter<SegmentComponent> {
vec!(
SegmentComponent::INFO,
/// Iterates through the components.
pub fn iterator() -> impl Iterator<Item = &'static SegmentComponent> {
static SEGMENT_COMPONENTS: [SegmentComponent; 7] = [
SegmentComponent::POSTINGS,
SegmentComponent::POSITIONS,
SegmentComponent::FASTFIELDS,
SegmentComponent::FIELDNORMS,
SegmentComponent::TERMS,
SegmentComponent::STORE,
).into_iter()
}
pub fn path_suffix(&self)-> &'static str {
match *self {
SegmentComponent::POSITIONS => ".pos",
SegmentComponent::INFO => ".info",
SegmentComponent::POSTINGS => ".idx",
SegmentComponent::TERMS => ".term",
SegmentComponent::STORE => ".store",
SegmentComponent::FASTFIELDS => ".fast",
SegmentComponent::FIELDNORMS => ".fieldnorm",
}
SegmentComponent::DELETE,
];
SEGMENT_COMPONENTS.into_iter()
}
}

59
src/core/segment_id.rs
View File

@@ -1,34 +1,36 @@
use uuid::Uuid;
use std::fmt;
use rustc_serialize::{Encoder, Decoder, Encodable, Decodable};
use core::SegmentComponent;
use std::path::PathBuf;
use std::cmp::{Ordering, Ord};
use std::cmp::{Ord, Ordering};
#[cfg(test)]
use std::sync::atomic;
#[derive(Clone, Copy, PartialEq, Eq, Hash)]
/// Uuid identifying a segment.
///
/// Tantivy's segment are identified
/// by a UUID which is used to prefix the filenames
/// of all of the file associated with the segment.
///
/// In unit test, for reproducability, the `SegmentId` are
/// simply generated in an autoincrement fashion.
#[derive(Clone, Copy, PartialEq, Eq, Hash, Serialize, Deserialize)]
pub struct SegmentId(Uuid);
#[cfg(test)]
lazy_static! {
static ref AUTO_INC_COUNTER: atomic::AtomicUsize = atomic::AtomicUsize::default();
static ref EMPTY_ARR: [u8; 8] = [0u8; 8];
}
// During tests, we generate the segment id in a autoincrement manner
// for consistency of segment id between run.
//
// The order of the test execution is not guaranteed, but the order
// The order of the test execution is not guaranteed, but the order
// of segments within a single test is guaranteed.
#[cfg(test)]
fn create_uuid() -> Uuid {
let new_auto_inc_id = (*AUTO_INC_COUNTER).fetch_add(1, atomic::Ordering::SeqCst);
Uuid::from_fields(new_auto_inc_id as u32, 0, 0, &*EMPTY_ARR)
Uuid::from_fields(new_auto_inc_id as u32, 0, 0, &*EMPTY_ARR).unwrap()
}
#[cfg(not(test))]
@@ -37,43 +39,34 @@ fn create_uuid() -> Uuid {
}
impl SegmentId {
#[doc(hidden)]
pub fn generate_random() -> SegmentId {
SegmentId(create_uuid())
}
pub fn short_uuid_string(&self,) -> String {
(&self.0.to_simple_string()[..8]).to_string()
}
pub fn uuid_string(&self,) -> String {
self.0.to_simple_string()
}
pub fn relative_path(&self, component: SegmentComponent) -> PathBuf {
let filename = self.uuid_string() + component.path_suffix();
PathBuf::from(filename)
}
}
impl Encodable for SegmentId {
fn encode<S: Encoder>(&self, s: &mut S) -> Result<(), S::Error> {
self.0.encode(s)
/// Returns a shorter identifier of the segment.
///
/// We are using UUID4, so only 6 bits are fixed,
/// and the rest is random.
///
/// Picking the first 8 chars is ok to identify
/// segments in a display message.
pub fn short_uuid_string(&self) -> String {
(&self.0.simple().to_string()[..8]).to_string()
}
}
impl Decodable for SegmentId {
fn decode<D: Decoder>(d: &mut D) -> Result<Self, D::Error> {
Uuid::decode(d).map(SegmentId)
/// Returns a segment uuid string.
pub fn uuid_string(&self) -> String {
self.0.simple().to_string()
}
}
impl fmt::Debug for SegmentId {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(f, "SegmentId({:?})", self.uuid_string())
write!(f, "Seg({:?})", self.short_uuid_string())
}
}
impl PartialOrd for SegmentId {
fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
Some(self.cmp(other))

115
src/core/segment_meta.rs Normal file
View File

@@ -0,0 +1,115 @@
use core::SegmentId;
use super::SegmentComponent;
use std::path::PathBuf;
use std::collections::HashSet;
#[derive(Clone, Debug, Serialize, Deserialize)]
struct DeleteMeta {
num_deleted_docs: u32,
opstamp: u64,
}
/// `SegmentMeta` contains simple meta information about a segment.
///
/// For instance the number of docs it contains,
/// how many are deleted, etc.
#[derive(Clone, Debug, Serialize, Deserialize)]
pub struct SegmentMeta {
segment_id: SegmentId,
max_doc: u32,
deletes: Option<DeleteMeta>,
}
impl SegmentMeta {
/// Creates a new segment meta for
/// a segment with no deletes and no documents.
pub fn new(segment_id: SegmentId) -> SegmentMeta {
SegmentMeta {
segment_id,
max_doc: 0,
deletes: None,
}
}
/// Returns the segment id.
pub fn id(&self) -> SegmentId {
self.segment_id
}
/// Returns the number of deleted documents.
pub fn num_deleted_docs(&self) -> u32 {
self.deletes
.as_ref()
.map(|delete_meta| delete_meta.num_deleted_docs)
.unwrap_or(0u32)
}
/// Returns the list of files that
/// are required for the segment meta.
///
/// This is useful as the way tantivy removes files
/// is by removing all files that have been created by tantivy
/// and are not used by any segment anymore.
pub fn list_files(&self) -> HashSet<PathBuf> {
SegmentComponent::iterator()
.map(|component| self.relative_path(*component))
.collect::<HashSet<PathBuf>>()
}
/// Returns the relative path of a component of our segment.
///
/// It just joins the segment id with the extension
/// associated to a segment component.
pub fn relative_path(&self, component: SegmentComponent) -> PathBuf {
let mut path = self.id().uuid_string();
path.push_str(&*match component {
SegmentComponent::POSITIONS => ".pos".to_string(),
SegmentComponent::POSTINGS => ".idx".to_string(),
SegmentComponent::TERMS => ".term".to_string(),
SegmentComponent::STORE => ".store".to_string(),
SegmentComponent::FASTFIELDS => ".fast".to_string(),
SegmentComponent::FIELDNORMS => ".fieldnorm".to_string(),
SegmentComponent::DELETE => format!(".{}.del", self.delete_opstamp().unwrap_or(0)),
});
PathBuf::from(path)
}
/// Return the highest doc id + 1
///
/// If there are no deletes, then num_docs = max_docs
/// and all the doc ids contains in this segment
/// are exactly (0..max_doc).
pub fn max_doc(&self) -> u32 {
self.max_doc
}
/// Return the number of documents in the segment.
pub fn num_docs(&self) -> u32 {
self.max_doc() - self.num_deleted_docs()
}
/// Returns the opstamp of the last delete operation
/// taken in account in this segment.
pub fn delete_opstamp(&self) -> Option<u64> {
self.deletes.as_ref().map(|delete_meta| delete_meta.opstamp)
}
/// Returns true iff the segment meta contains
/// delete information.
pub fn has_deletes(&self) -> bool {
self.deletes.is_some()
}
#[doc(hidden)]
pub fn set_max_doc(&mut self, max_doc: u32) {
self.max_doc = max_doc;
}
#[doc(hidden)]
pub fn set_delete_meta(&mut self, num_deleted_docs: u32, opstamp: u64) {
self.deletes = Some(DeleteMeta {
num_deleted_docs,
opstamp,
});
}
}

327
src/core/segment_reader.rs
View File

@@ -2,28 +2,23 @@ use Result;
use core::Segment;
use core::SegmentId;
use core::SegmentComponent;
use schema::Term;
use std::sync::RwLock;
use common::HasLen;
use core::SegmentMeta;
use fastfield::{self, FastFieldNotAvailableError};
use fastfield::DeleteBitSet;
use store::StoreReader;
use schema::Document;
use directory::ReadOnlySource;
use schema::Document;
use DocId;
use std::io;
use std::str;
use postings::TermInfo;
use datastruct::FstMap;
use std::sync::Arc;
use std::collections::HashMap;
use common::CompositeFile;
use std::fmt;
use rustc_serialize::json;
use core::SegmentInfo;
use core::InvertedIndexReader;
use schema::Field;
use postings::SegmentPostingsOption;
use postings::SegmentPostings;
use fastfield::{U32FastFieldsReader, U32FastFieldReader};
use fastfield::{FastFieldReader, U64FastFieldReader};
use schema::Schema;
use schema::FieldType;
use postings::FreqHandler;
use schema::TextIndexingOptions;
use error::Error;
/// Entry point to access all of the datastructures of the `Segment`
///
@@ -36,15 +31,21 @@ use error::Error;
/// The segment reader has a very low memory footprint,
/// as close to all of the memory data is mmapped.
///
#[derive(Clone)]
pub struct SegmentReader {
segment_info: SegmentInfo,
inv_idx_reader_cache: Arc<RwLock<HashMap<Field, Arc<InvertedIndexReader>>>>,
segment_id: SegmentId,
term_infos: FstMap<TermInfo>,
postings_data: ReadOnlySource,
segment_meta: SegmentMeta,
termdict_composite: CompositeFile,
postings_composite: CompositeFile,
positions_composite: CompositeFile,
fast_fields_composite: CompositeFile,
fieldnorms_composite: CompositeFile,
store_reader: StoreReader,
fast_fields_reader: U32FastFieldsReader,
fieldnorms_reader: U32FastFieldsReader,
positions_data: ReadOnlySource,
delete_bitset: DeleteBitSet,
schema: Schema,
}
@@ -54,108 +55,160 @@ impl SegmentReader {
/// Today, `tantivy` does not handle deletes, so it happens
/// to also be the number of documents in the index.
pub fn max_doc(&self) -> DocId {
self.segment_info.max_doc
self.segment_meta.max_doc()
}
/// Returns the number of documents.
/// Deleted documents are not counted.
///
/// Today, `tantivy` does not handle deletes so max doc and
/// num_docs are the same.
pub fn num_docs(&self) -> DocId {
self.segment_info.max_doc
self.segment_meta.num_docs()
}
/// Return the number of documents that have been
/// deleted in the segment.
pub fn num_deleted_docs(&self) -> DocId {
self.delete_bitset.len() as DocId
}
/// Accessor to a segment's fast field reader given a field.
pub fn get_fast_field_reader(&self, field: Field) -> io::Result<U32FastFieldReader> {
///
/// Returns the u64 fast value reader if the field
/// is a u64 field indexed as "fast".
///
/// Return a FastFieldNotAvailableError if the field is not
/// declared as a fast field in the schema.
///
/// # Panics
/// May panic if the index is corrupted.
pub fn get_fast_field_reader<TFastFieldReader: FastFieldReader>(
&self,
field: Field,
) -> fastfield::Result<TFastFieldReader> {
let field_entry = self.schema.get_field_entry(field);
match *field_entry.field_type() {
FieldType::Str(_) => {
Err(io::Error::new(io::ErrorKind::Other, "fast field are not yet supported for text fields."))
},
FieldType::U32(_) => {
// TODO check that the schema allows that
//Err(io::Error::new(io::ErrorKind::Other, "fast field are not yet supported for text fields."))
self.fast_fields_reader.get_field(field)
},
if !TFastFieldReader::is_enabled(field_entry.field_type()) {
Err(FastFieldNotAvailableError::new(field_entry))
} else {
self.fast_fields_composite
.open_read(field)
.ok_or_else(|| FastFieldNotAvailableError::new(field_entry))
.map(TFastFieldReader::open)
}
}
/// Accessor to the segment's `Field norms`'s reader.
///
/// Field norms are the length (in tokens) of the fields.
/// It is used in the computation of the [TfIdf](https://fulmicoton.gitbooks.io/tantivy-doc/content/tfidf.html).
/// It is used in the computation of the [TfIdf]
/// (https://fulmicoton.gitbooks.io/tantivy-doc/content/tfidf.html).
///
/// They are simply stored as a fast field, serialized in
/// the `.fieldnorm` file of the segment.
pub fn get_fieldnorms_reader(&self, field: Field) -> io::Result<U32FastFieldReader> {
self.fieldnorms_reader.get_field(field)
/// They are simply stored as a fast field, serialized in
/// the `.fieldnorm` file of the segment.
pub fn get_fieldnorms_reader(&self, field: Field) -> Option<U64FastFieldReader> {
self.fieldnorms_composite
.open_read(field)
.map(U64FastFieldReader::open)
}
/// Returns the number of documents containing the term.
pub fn doc_freq(&self, term: &Term) -> u32 {
match self.get_term_info(term) {
Some(term_info) => term_info.doc_freq,
None => 0,
}
}
/// Accessor to the segment's `StoreReader`.
pub fn get_store_reader(&self) -> &StoreReader {
&self.store_reader
}
/// Open a new segment for reading.
pub fn open(segment: Segment) -> Result<SegmentReader> {
let segment_info_reader = try!(segment.open_read(SegmentComponent::INFO));
let segment_info_data = try!(
str::from_utf8(&*segment_info_reader)
.map_err(|err| {
let segment_info_filepath = segment.relative_path(SegmentComponent::INFO);
Error::CorruptedFile(segment_info_filepath, Box::new(err))
})
);
let segment_info: SegmentInfo = try!(
json::decode(&segment_info_data)
.map_err(|err| {
let file_path = segment.relative_path(SegmentComponent::INFO);
Error::CorruptedFile(file_path, Box::new(err))
})
);
let source = try!(segment.open_read(SegmentComponent::TERMS));
let term_infos = try!(FstMap::from_source(source));
let store_reader = StoreReader::from(try!(segment.open_read(SegmentComponent::STORE)));
let postings_shared_mmap = try!(segment.open_read(SegmentComponent::POSTINGS));
let fast_field_data = try!(segment.open_read(SegmentComponent::FASTFIELDS));
let fast_fields_reader = try!(U32FastFieldsReader::open(fast_field_data));
let fieldnorms_data = try!(segment.open_read(SegmentComponent::FIELDNORMS));
let fieldnorms_reader = try!(U32FastFieldsReader::open(fieldnorms_data));
let positions_data = segment
.open_read(SegmentComponent::POSITIONS)
.unwrap_or_else(|_| ReadOnlySource::empty());
pub fn open(segment: &Segment) -> Result<SegmentReader> {
let termdict_source = segment.open_read(SegmentComponent::TERMS)?;
let termdict_composite = CompositeFile::open(&termdict_source)?;
let store_source = segment.open_read(SegmentComponent::STORE)?;
let store_reader = StoreReader::from_source(store_source);
let postings_source = segment.open_read(SegmentComponent::POSTINGS)?;
let postings_composite = CompositeFile::open(&postings_source)?;
let positions_composite = {
if let Ok(source) = segment.open_read(SegmentComponent::POSITIONS) {
CompositeFile::open(&source)?
} else {
CompositeFile::empty()
}
};
let fast_fields_data = segment.open_read(SegmentComponent::FASTFIELDS)?;
let fast_fields_composite = CompositeFile::open(&fast_fields_data)?;
let fieldnorms_data = segment.open_read(SegmentComponent::FIELDNORMS)?;
let fieldnorms_composite = CompositeFile::open(&fieldnorms_data)?;
let delete_bitset = if segment.meta().has_deletes() {
let delete_data = segment.open_read(SegmentComponent::DELETE)?;
DeleteBitSet::open(delete_data)
} else {
DeleteBitSet::empty()
};
let schema = segment.schema();
Ok(SegmentReader {
segment_info: segment_info,
postings_data: postings_shared_mmap,
term_infos: term_infos,
inv_idx_reader_cache: Arc::new(RwLock::new(HashMap::new())),
segment_meta: segment.meta().clone(),
termdict_composite,
postings_composite,
fast_fields_composite,
fieldnorms_composite,
segment_id: segment.id(),
store_reader: store_reader,
fast_fields_reader: fast_fields_reader,
fieldnorms_reader: fieldnorms_reader,
positions_data: positions_data,
schema: schema,
store_reader,
delete_bitset,
positions_composite,
schema,
})
}
/// Return the term dictionary datastructure.
pub fn term_infos(&self) -> &FstMap<TermInfo> {
&self.term_infos
/// Returns a field reader associated to the field given in argument.
///
/// The field reader is in charge of iterating through the
/// term dictionary associated to a specific field,
/// and opening the posting list associated to any term.
pub fn inverted_index(&self, field: Field) -> Arc<InvertedIndexReader> {
if let Some(inv_idx_reader) = self.inv_idx_reader_cache
.read()
.expect("Lock poisoned. This should never happen")
.get(&field)
{
Arc::clone(inv_idx_reader);
}
let termdict_source: ReadOnlySource = self.termdict_composite
.open_read(field)
.expect("Index corrupted. Failed to open field term dictionary in composite file.");
let postings_source = self.postings_composite
.open_read(field)
.expect("Index corrupted. Failed to open field postings in composite file.");
let positions_source = self.positions_composite
.open_read(field)
.expect("Index corrupted. Failed to open field positions in composite file.");
let inv_idx_reader = Arc::new(InvertedIndexReader::new(
termdict_source,
postings_source,
positions_source,
self.delete_bitset.clone(),
self.schema.clone(),
));
// by releasing the lock in between, we may end up opening the inverting index
// twice, but this is fine.
self.inv_idx_reader_cache
.write()
.expect("Field reader cache lock poisoned. This should never happen.")
.insert(field, Arc::clone(&inv_idx_reader));
inv_idx_reader
}
/// Returns the document (or to be accurate, its stored field)
/// bearing the given doc id.
/// This method is slow and should seldom be called from
@@ -164,82 +217,24 @@ impl SegmentReader {
self.store_reader.get(doc_id)
}
/// Returns the segment id
pub fn segment_id(&self) -> SegmentId {
self.segment_id
}
/// Returns the segment postings associated with the term, and with the given option,
/// or `None` if the term has never been encounterred and indexed.
///
/// If the field was not indexed with the indexing options that cover
/// the requested options, the returned `SegmentPostings` the method does not fail
/// and returns a `SegmentPostings` with as much information as possible.
///
/// For instance, requesting `SegmentPostingsOption::FreqAndPositions` for a `TextIndexingOptions`
/// that does not index position will return a `SegmentPostings` with `DocId`s and frequencies.
pub fn read_postings(&self, term: &Term, option: SegmentPostingsOption) -> Option<SegmentPostings> {
let field = term.field();
let field_entry = self.schema.get_field_entry(field);
let term_info = get!(self.get_term_info(&term));
let offset = term_info.postings_offset as usize;
let postings_data = &self.postings_data[offset..];
let freq_handler = match *field_entry.field_type() {
FieldType::Str(ref options) => {
let indexing_options = options.get_indexing_options();
match option {
SegmentPostingsOption::NoFreq => {
FreqHandler::new_without_freq()
}
SegmentPostingsOption::Freq => {
if indexing_options.is_termfreq_enabled() {
FreqHandler::new_with_freq()
}
else {
FreqHandler::new_without_freq()
}
}
SegmentPostingsOption::FreqAndPositions => {
if indexing_options == TextIndexingOptions::TokenizedWithFreqAndPosition {
let offseted_position_data = &self.positions_data[term_info.positions_offset as usize ..];
FreqHandler::new_with_freq_and_position(offseted_position_data)
}
else if indexing_options.is_termfreq_enabled()
{
FreqHandler::new_with_freq()
}
else {
FreqHandler::new_without_freq()
}
}
}
}
_ => {
FreqHandler::new_without_freq()
}
};
Some(SegmentPostings::from_data(term_info.doc_freq, postings_data, freq_handler))
/// Returns the bitset representing
/// the documents that have been deleted.
pub fn delete_bitset(&self) -> &DeleteBitSet {
&self.delete_bitset
}
/// Returns the posting list associated with a term.
pub fn read_postings_all_info(&self, term: &Term) -> Option<SegmentPostings> {
let field_entry = self.schema.get_field_entry(term.field());
let segment_posting_option = match *field_entry.field_type() {
FieldType::Str(ref text_options) => {
match text_options.get_indexing_options() {
TextIndexingOptions::TokenizedWithFreq => SegmentPostingsOption::Freq,
TextIndexingOptions::TokenizedWithFreqAndPosition => SegmentPostingsOption::FreqAndPositions,
_ => SegmentPostingsOption::NoFreq,
}
}
FieldType::U32(_) => SegmentPostingsOption::NoFreq
};
self.read_postings(term, segment_posting_option)
}
/// Returns the term info associated with the term.
pub fn get_term_info(&self, term: &Term) -> Option<TermInfo> {
self.term_infos.get(term.as_slice())
/// Returns true iff the `doc` is marked
/// as deleted.
pub fn is_deleted(&self, doc: DocId) -> bool {
self.delete_bitset.is_deleted(doc)
}
}
impl fmt::Debug for SegmentReader {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(f, "SegmentReader({:?})", self.segment_id)

149
src/datastruct/fstmap.rs
View File

@@ -1,149 +0,0 @@
#![allow(should_implement_trait)]
use std::io;
use std::io::Write;
use fst;
use fst::raw::Fst;
use fst::Streamer;
use directory::ReadOnlySource;
use common::BinarySerializable;
use std::marker::PhantomData;
fn convert_fst_error(e: fst::Error) -> io::Error {
io::Error::new(io::ErrorKind::Other, e)
}
pub struct FstMapBuilder<W: Write, V: BinarySerializable> {
fst_builder: fst::MapBuilder<W>,
data: Vec<u8>,
_phantom_: PhantomData<V>,
}
impl<W: Write, V: BinarySerializable> FstMapBuilder<W, V> {
pub fn new(w: W) -> io::Result<FstMapBuilder<W, V>> {
let fst_builder = try!(fst::MapBuilder::new(w).map_err(convert_fst_error));
Ok(FstMapBuilder {
fst_builder: fst_builder,
data: Vec::new(),
_phantom_: PhantomData,
})
}
pub fn insert(&mut self, key: &[u8], value: &V) -> io::Result<()>{
try!(self.fst_builder
.insert(key, self.data.len() as u64)
.map_err(convert_fst_error));
try!(value.serialize(&mut self.data));
Ok(())
}
pub fn finish(self,) -> io::Result<W> {
let mut file = try!(
self.fst_builder
.into_inner()
.map_err(convert_fst_error));
let footer_size = self.data.len() as u32;
try!(file.write_all(&self.data));
try!((footer_size as u32).serialize(&mut file));
try!(file.flush());
Ok(file)
}
}
pub struct FstMap<V: BinarySerializable> {
fst_index: fst::Map,
values_mmap: ReadOnlySource,
_phantom_: PhantomData<V>,
}
fn open_fst_index(source: ReadOnlySource) -> io::Result<fst::Map> {
Ok(fst::Map::from(match source {
ReadOnlySource::Anonymous(data) => try!(Fst::from_shared_bytes(data.data, data.start, data.len).map_err(convert_fst_error)),
ReadOnlySource::Mmap(mmap_readonly) => try!(Fst::from_mmap(mmap_readonly).map_err(convert_fst_error)),
}))
}
pub struct FstKeyIter<'a, V: 'static + BinarySerializable> {
streamer: fst::map::Stream<'a>,
__phantom__: PhantomData<V>
}
impl<'a, V: 'static + BinarySerializable> FstKeyIter<'a, V> {
pub fn next(&mut self) -> Option<(&[u8])> {
self.streamer
.next()
.map(|(k, _)| k)
}
}
impl<V: BinarySerializable> FstMap<V> {
pub fn keys(&self,) -> FstKeyIter<V> {
FstKeyIter {
streamer: self.fst_index.stream(),
__phantom__: PhantomData,
}
}
pub fn from_source(source: ReadOnlySource) -> io::Result<FstMap<V>> {
let total_len = source.len();
let length_offset = total_len - 4;
let mut split_len_buffer: &[u8] = &source.as_slice()[length_offset..];
let footer_size = try!(u32::deserialize(&mut split_len_buffer)) as usize;
let split_len = length_offset - footer_size;
let fst_source = source.slice(0, split_len);
let values_source = source.slice(split_len, length_offset);
let fst_index = try!(open_fst_index(fst_source));
Ok(FstMap {
fst_index: fst_index,
values_mmap: values_source,
_phantom_: PhantomData,
})
}
fn read_value(&self, offset: u64) -> V {
let buffer = self.values_mmap.as_slice();
let mut cursor = &buffer[(offset as usize)..];
V::deserialize(&mut cursor).expect("Data in FST is corrupted")
}
pub fn get<K: AsRef<[u8]>>(&self, key: K) -> Option<V> {
self.fst_index
.get(key)
.map(|offset| self.read_value(offset))
}
}
#[cfg(test)]
mod tests {
use super::*;
use directory::{RAMDirectory, Directory};
use std::path::PathBuf;
#[test]
fn test_fstmap() {
let mut directory = RAMDirectory::create();
let path = PathBuf::from("fstmap");
{
let write = directory.open_write(&path).unwrap();
let mut fstmap_builder = FstMapBuilder::new(write).unwrap();
fstmap_builder.insert("abc".as_bytes(), &34u32).unwrap();
fstmap_builder.insert("abcd".as_bytes(), &346u32).unwrap();
fstmap_builder.finish().unwrap();
}
let source = directory.open_read(&path).unwrap();
let fstmap: FstMap<u32> = FstMap::from_source(source).unwrap();
assert_eq!(fstmap.get("abc"), Some(34u32));
assert_eq!(fstmap.get("abcd"), Some(346u32));
let mut keys = fstmap.keys();
assert_eq!(keys.next().unwrap(), "abc".as_bytes());
assert_eq!(keys.next().unwrap(), "abcd".as_bytes());
assert_eq!(keys.next(), None);
}
}

6
src/datastruct/mod.rs
View File

@@ -1,8 +1,4 @@
mod fstmap;
mod skip;
pub mod stacker;
pub use self::fstmap::FstMapBuilder;
pub use self::fstmap::FstMap;
pub use self::fstmap::FstKeyIter;
pub use self::skip::{SkipListBuilder, SkipList};
pub use self::skip::{SkipList, SkipListBuilder};

6
src/datastruct/skip/mod.rs
View File

@@ -6,8 +6,6 @@ mod skiplist;
pub use self::skiplist_builder::SkipListBuilder;
pub use self::skiplist::SkipList;
#[cfg(test)]
mod tests {
@@ -114,9 +112,9 @@ mod tests {
let mut skip_list: SkipList<()> = SkipList::from(output.as_slice());
assert_eq!(skip_list.next().unwrap(), (0, ()));
skip_list.seek(431);
assert_eq!(skip_list.next().unwrap(), (431,()) );
assert_eq!(skip_list.next().unwrap(), (431, ()));
skip_list.seek(1003);
assert_eq!(skip_list.next().unwrap(), (1004,()) );
assert_eq!(skip_list.next().unwrap(), (1004, ()));
assert_eq!(skip_list.next(), None);
}

67
src/datastruct/skip/skiplist.rs
View File

@@ -13,14 +13,12 @@ struct Layer<'a, T> {
}
impl<'a, T: BinarySerializable> Iterator for Layer<'a, T> {
type Item = (DocId, T);
fn next(&mut self,)-> Option<(DocId, T)> {
fn next(&mut self) -> Option<(DocId, T)> {
if self.next_id == u32::max_value() {
None
}
else {
} else {
let cur_val = T::deserialize(&mut self.cursor).unwrap();
let cur_id = self.next_id;
self.next_id = u32::deserialize(&mut self.cursor).unwrap_or(u32::max_value());
@@ -31,19 +29,18 @@ impl<'a, T: BinarySerializable> Iterator for Layer<'a, T> {
impl<'a, T: BinarySerializable> From<&'a [u8]> for Layer<'a, T> {
fn from(data: &'a [u8]) -> Layer<'a, T> {
let mut cursor = data;
let mut cursor = data;
let next_id = u32::deserialize(&mut cursor).unwrap_or(u32::max_value());
Layer {
data: data,
cursor: cursor,
next_id: next_id,
data,
cursor,
next_id,
_phantom_: PhantomData,
}
}
}
impl<'a, T: BinarySerializable> Layer<'a, T> {
fn empty() -> Layer<'a, T> {
Layer {
data: &EMPTY,
@@ -53,11 +50,11 @@ impl<'a, T: BinarySerializable> Layer<'a, T> {
}
}
fn seek_offset(&mut self, offset: usize) {
fn seek_offset(&mut self, offset: usize) {
self.cursor = &self.data[offset..];
self.next_id = u32::deserialize(&mut self.cursor).unwrap_or(u32::max_value());
}
// Returns the last element (key, val)
// such that (key < doc_id)
//
@@ -67,31 +64,32 @@ impl<'a, T: BinarySerializable> Layer<'a, T> {
let mut val = None;
while self.next_id < doc_id {
match self.next() {
None => { break; },
v => { val = v; }
None => {
break;
}
v => {
val = v;
}
}
}
val
}
}
pub struct SkipList<'a, T: BinarySerializable> {
data_layer: Layer<'a, T>,
skip_layers: Vec<Layer<'a, u32>>,
}
impl<'a, T: BinarySerializable> Iterator for SkipList<'a, T> {
type Item = (DocId, T);
fn next(&mut self,)-> Option<(DocId, T)> {
fn next(&mut self) -> Option<(DocId, T)> {
self.data_layer.next()
}
}
impl<'a, T: BinarySerializable> SkipList<'a, T> {
pub fn seek(&mut self, doc_id: DocId) -> Option<(DocId, T)> {
let mut next_layer_skip: Option<(DocId, u32)> = None;
for skip_layer in &mut self.skip_layers {
@@ -99,39 +97,32 @@ impl<'a, T: BinarySerializable> SkipList<'a, T> {
skip_layer.seek_offset(offset as usize);
}
next_layer_skip = skip_layer.seek(doc_id);
}
if let Some((_, offset)) = next_layer_skip {
self.data_layer.seek_offset(offset as usize);
}
self.data_layer.seek(doc_id)
}
if let Some((_, offset)) = next_layer_skip {
self.data_layer.seek_offset(offset as usize);
}
self.data_layer.seek(doc_id)
}
}
impl<'a, T: BinarySerializable> From<&'a [u8]> for SkipList<'a, T> {
fn from(mut data: &'a [u8]) -> SkipList<'a, T> {
let offsets: Vec<u32> = Vec::deserialize(&mut data).unwrap();
let num_layers = offsets.len();
let layers_data: &[u8] = data;
let data_layer: Layer<'a, T> =
if num_layers == 0 { Layer::empty() }
else {
let first_layer_data: &[u8] = &layers_data[..offsets[0] as usize];
Layer::from(first_layer_data)
};
let data_layer: Layer<'a, T> = if num_layers == 0 {
Layer::empty()
} else {
let first_layer_data: &[u8] = &layers_data[..offsets[0] as usize];
Layer::from(first_layer_data)
};
let skip_layers = (0..max(1, num_layers) - 1)
.map(|i| (offsets[i] as usize, offsets[i + 1] as usize))
.map(|(start, stop)| {
Layer::from(&layers_data[start..stop])
})
.map(|(start, stop)| Layer::from(&layers_data[start..stop]))
.collect();
SkipList {
skip_layers: skip_layers,
data_layer: data_layer,
skip_layers,
data_layer,
}
}
}

47
src/datastruct/skip/skiplist_builder.rs
View File

@@ -13,19 +13,18 @@ struct LayerBuilder<T: BinarySerializable> {
}
impl<T: BinarySerializable> LayerBuilder<T> {
fn written_size(&self,) -> usize {
fn written_size(&self) -> usize {
self.buffer.len()
}
fn write(&self, output: &mut Write) -> Result<(), io::Error> {
try!(output.write_all(&self.buffer));
output.write_all(&self.buffer)?;
Ok(())
}
fn with_period(period: usize) -> LayerBuilder<T> {
LayerBuilder {
period: period,
period,
buffer: Vec::new(),
remaining: period,
len: 0,
@@ -36,30 +35,28 @@ impl<T: BinarySerializable> LayerBuilder<T> {
fn insert(&mut self, doc_id: DocId, value: &T) -> io::Result<Option<(DocId, u32)>> {
self.remaining -= 1;
self.len += 1;
let offset = self.written_size() as u32; // TODO not sure if we want after or here
try!(doc_id.serialize(&mut self.buffer));
try!(value.serialize(&mut self.buffer));
let offset = self.written_size() as u32;
doc_id.serialize(&mut self.buffer)?;
value.serialize(&mut self.buffer)?;
Ok(if self.remaining == 0 {
self.remaining = self.period;
Some((doc_id, offset))
}
else { None })
self.remaining = self.period;
Some((doc_id, offset))
} else {
None
})
}
}
pub struct SkipListBuilder<T: BinarySerializable> {
period: usize,
data_layer: LayerBuilder<T>,
skip_layers: Vec<LayerBuilder<u32>>,
}
impl<T: BinarySerializable> SkipListBuilder<T> {
pub fn new(period: usize) -> SkipListBuilder<T> {
SkipListBuilder {
period: period,
period,
data_layer: LayerBuilder::with_period(period),
skip_layers: Vec::new(),
}
@@ -75,20 +72,20 @@ impl<T: BinarySerializable> SkipListBuilder<T> {
pub fn insert(&mut self, doc_id: DocId, dest: &T) -> io::Result<()> {
let mut layer_id = 0;
let mut skip_pointer = try!(self.data_layer.insert(doc_id, dest));
let mut skip_pointer = self.data_layer.insert(doc_id, dest)?;
loop {
skip_pointer = match skip_pointer {
Some((skip_doc_id, skip_offset)) =>
try!(self
.get_skip_layer(layer_id)
.insert(skip_doc_id, &skip_offset)),
None => { return Ok(()); }
Some((skip_doc_id, skip_offset)) => self.get_skip_layer(layer_id)
.insert(skip_doc_id, &skip_offset)?,
None => {
return Ok(());
}
};
layer_id += 1;
}
}
pub fn write<W: Write>(self, output: &mut Write) -> io::Result<()> {
pub fn write<W: Write>(self, output: &mut W) -> io::Result<()> {
let mut size: u32 = 0;
let mut layer_sizes: Vec<u32> = Vec::new();
size += self.data_layer.buffer.len() as u32;
@@ -97,10 +94,10 @@ impl<T: BinarySerializable> SkipListBuilder<T> {
size += layer.buffer.len() as u32;
layer_sizes.push(size);
}
try!(layer_sizes.serialize(output));
try!(self.data_layer.write(output));
layer_sizes.serialize(output)?;
self.data_layer.write(output)?;
for layer in self.skip_layers.iter().rev() {
try!(layer.write(output));
layer.write(output)?;
}
Ok(())
}

39
src/datastruct/stacker/expull.rs
View File

@@ -1,7 +1,6 @@
use std::mem;
use super::heap::{Heap, HeapAllocable};
#[inline]
pub fn is_power_of_2(val: u32) -> bool {
val & (val - 1) == 0
@@ -9,11 +8,10 @@ pub fn is_power_of_2(val: u32) -> bool {
#[inline]
pub fn jump_needed(val: u32) -> bool {
val > 3 && is_power_of_2(val)
val > 3 && is_power_of_2(val)
}
#[derive(Debug)]
#[derive(Debug, Clone)]
pub struct ExpUnrolledLinkedList {
len: u32,
end: u32,
@@ -24,10 +22,9 @@ pub struct ExpUnrolledLinkedList {
}
impl ExpUnrolledLinkedList {
pub fn iter<'a>(&self, addr: u32, heap: &'a Heap) -> ExpUnrolledLinkedListIterator<'a> {
ExpUnrolledLinkedListIterator {
heap: heap,
heap,
addr: addr + 2u32 * (mem::size_of::<u32>() as u32),
len: self.len,
consumed: 0,
@@ -42,16 +39,21 @@ impl ExpUnrolledLinkedList {
// the next block as a size of (length so far),
// and we need to add 1u32 to store the pointer
// to the next element.
let new_block_size: usize = (self.len as usize + 1) * mem::size_of::<u32>();
let new_block_size: usize = (self.len as usize + 1) * mem::size_of::<u32>();
let new_block_addr: u32 = heap.allocate_space(new_block_size);
heap.set(self.end, &new_block_addr);
self.end = new_block_addr;
self.end = new_block_addr;
}
heap.set(self.end, &val);
self.end += mem::size_of::<u32>() as u32;
}
}
impl HeapAllocable for u32 {
fn with_addr(_addr: u32) -> u32 {
0u32
}
}
impl HeapAllocable for ExpUnrolledLinkedList {
fn with_addr(addr: u32) -> ExpUnrolledLinkedList {
@@ -77,33 +79,26 @@ pub struct ExpUnrolledLinkedListIterator<'a> {
impl<'a> Iterator for ExpUnrolledLinkedListIterator<'a> {
type Item = u32;
fn next(&mut self,) -> Option<u32> {
fn next(&mut self) -> Option<u32> {
if self.consumed == self.len {
None
}
else {
} else {
let addr: u32;
self.consumed += 1;
if jump_needed(self.consumed) {
addr = *self.heap.get_mut_ref(self.addr);
}
else {
} else {
addr = self.addr;
}
self.addr = addr + mem::size_of::<u32>() as u32;
Some(*self.heap.get_mut_ref(addr))
}
}
}
}
#[cfg(test)]
mod tests {
use super::*;
use super::super::heap::Heap;
use test::Bencher;
@@ -147,7 +142,7 @@ mod tests {
#[bench]
fn bench_push_stack(bench: &mut Bencher) {
let heap = Heap::with_capacity(64_000_000);
let heap = Heap::with_capacity(64_000_000);
bench.iter(|| {
let mut stacks = Vec::with_capacity(100);
for _ in 0..NUM_STACK {
@@ -163,4 +158,4 @@ mod tests {
heap.clear();
});
}
}
}

301
src/datastruct/stacker/hashmap.rs
View File

@@ -1,26 +1,76 @@
use std::iter;
use std::marker::PhantomData;
use super::heap::{Heap, HeapAllocable, BytesRef};
use std::mem;
use super::heap::{BytesRef, Heap, HeapAllocable};
mod murmurhash2 {
const SEED: u32 = 3_242_157_231u32;
#[inline(always)]
pub fn murmurhash2(key: &[u8]) -> u32 {
let mut key_ptr: *const u32 = key.as_ptr() as *const u32;
let m: u32 = 0x5bd1_e995;
let r = 24;
let len = key.len() as u32;
/// dbj2 hash function
fn djb2(key: &[u8]) -> u64 {
let mut state: u64 = 5381;
for &b in key {
state = (state << 5).wrapping_add(state).wrapping_add(b as u64);
let mut h: u32 = SEED ^ len;
let num_blocks = len >> 2;
for _ in 0..num_blocks {
let mut k: u32 = unsafe { *key_ptr };
k = k.wrapping_mul(m);
k ^= k >> r;
k = k.wrapping_mul(m);
k = k.wrapping_mul(m);
h ^= k;
key_ptr = key_ptr.wrapping_offset(1);
}
// Handle the last few bytes of the input array
let remaining = len & 3;
let key_ptr_u8: *const u8 = key_ptr as *const u8;
match remaining {
3 => {
h ^= unsafe { u32::from(*key_ptr_u8.wrapping_offset(2)) } << 16;
h ^= unsafe { u32::from(*key_ptr_u8.wrapping_offset(1)) } << 8;
h ^= unsafe { u32::from(*key_ptr_u8) };
h = h.wrapping_mul(m);
}
2 => {
h ^= unsafe { u32::from(*key_ptr_u8.wrapping_offset(1)) } << 8;
h ^= unsafe { u32::from(*key_ptr_u8) };
h = h.wrapping_mul(m);
}
1 => {
h ^= unsafe { u32::from(*key_ptr_u8) };
h = h.wrapping_mul(m);
}
_ => {}
}
h ^= h >> 13;
h = h.wrapping_mul(m);
h ^ (h >> 15)
}
state
}
impl Default for BytesRef {
fn default() -> BytesRef {
BytesRef {
start: 0u32,
stop: 0u32,
}
}
/// Split the thread memory budget into
/// - the heap size
/// - the hash table "table" itself.
///
/// Returns (the heap size in bytes, the hash table size in number of bits)
pub(crate) fn split_memory(per_thread_memory_budget: usize) -> (usize, usize) {
let table_size_limit: usize = per_thread_memory_budget / 3;
let compute_table_size = |num_bits: usize| (1 << num_bits) * mem::size_of::<KeyValue>();
let table_num_bits: usize = (1..)
.into_iter()
.take_while(|num_bits: &usize| compute_table_size(*num_bits) < table_size_limit)
.last()
.expect(&format!(
"Per thread memory is too small: {}",
per_thread_memory_budget
));
let table_size = compute_table_size(table_num_bits);
let heap_size = per_thread_memory_budget - table_size;
(heap_size, table_num_bits)
}
/// `KeyValue` is the item stored in the hash table.
@@ -29,27 +79,21 @@ impl Default for BytesRef {
///
/// The key and the value are actually stored contiguously.
/// For this reason, the (start, stop) information is actually redundant
/// and can be simplified in the future
/// and can be simplified in the future
#[derive(Copy, Clone, Default)]
struct KeyValue {
key: BytesRef,
value_addr: u32,
key_value_addr: BytesRef,
hash: u32,
}
impl KeyValue {
fn is_empty(&self,) -> bool {
self.key.stop == 0u32
fn is_empty(&self) -> bool {
self.key_value_addr.is_null()
}
}
pub enum Entry {
Vacant(usize),
Occupied(u32),
}
/// Customized `HashMap` with string keys
///
///
/// This `HashMap` takes String as keys. Keys are
/// stored in a user defined heap.
///
@@ -57,114 +101,105 @@ pub enum Entry {
/// the computation of the hash of the key twice,
/// or copying the key as long as there is no insert.
///
pub struct HashMap<'a, V> where V: HeapAllocable {
pub struct HashMap<'a> {
table: Box<[KeyValue]>,
heap: &'a Heap,
_phantom: PhantomData<V>,
mask: usize,
occupied: Vec<usize>,
}
impl<'a, V> HashMap<'a, V> where V: HeapAllocable {
struct QuadraticProbing {
hash: usize,
i: usize,
mask: usize,
}
pub fn new(num_bucket_power_of_2: usize, heap: &'a Heap) -> HashMap<'a, V> {
impl QuadraticProbing {
fn compute(hash: usize, mask: usize) -> QuadraticProbing {
QuadraticProbing { hash, i: 0, mask }
}
#[inline]
fn next_probe(&mut self) -> usize {
self.i += 1;
(self.hash + self.i * self.i) & self.mask
}
}
impl<'a> HashMap<'a> {
pub fn new(num_bucket_power_of_2: usize, heap: &'a Heap) -> HashMap<'a> {
let table_size = 1 << num_bucket_power_of_2;
let table: Vec<KeyValue> = iter::repeat(KeyValue::default())
.take(table_size)
.collect();
let table: Vec<KeyValue> = iter::repeat(KeyValue::default()).take(table_size).collect();
HashMap {
table: table.into_boxed_slice(),
heap: heap,
_phantom: PhantomData,
heap,
mask: table_size - 1,
occupied: Vec::with_capacity(table_size / 2),
}
}
#[inline]
fn bucket(&self, key: &[u8]) -> usize {
let hash: u64 = djb2(key);
(hash as usize) & self.mask
fn probe(&self, hash: u32) -> QuadraticProbing {
QuadraticProbing::compute(hash as usize, self.mask)
}
fn get_key(&self, bytes_ref: BytesRef) -> &[u8] {
self.heap.get_slice(bytes_ref)
pub fn is_saturated(&self) -> bool {
self.table.len() < self.occupied.len() * 3
}
pub fn set_bucket(&mut self, key_bytes: &[u8], bucket: usize, addr: u32) -> u32 {
#[inline(never)]
fn get_key_value(&self, bytes_ref: BytesRef) -> (&[u8], u32) {
let key_bytes: &[u8] = self.heap.get_slice(bytes_ref);
let expull_addr: u32 = bytes_ref.addr() + 2 + key_bytes.len() as u32;
(key_bytes, expull_addr)
}
pub fn set_bucket(&mut self, hash: u32, key_bytes_ref: BytesRef, bucket: usize) {
self.occupied.push(bucket);
self.table[bucket] = KeyValue {
key: self.heap.allocate_and_set(key_bytes),
value_addr: addr,
key_value_addr: key_bytes_ref,
hash,
};
addr
}
pub fn iter<'b: 'a>(&'b self,) -> impl Iterator<Item=(&'a [u8], (u32, &'a V))> + 'b {
let heap: &'a Heap = self.heap;
let table: &'b [KeyValue] = &self.table;
self.occupied
.iter()
.cloned()
.map(move |bucket: usize| {
let kv = table[bucket];
let addr = kv.value_addr;
let v: &V = heap.get_mut_ref::<V>(addr);
(heap.get_slice(kv.key), (addr, v))
})
// .map(move |addr: u32| (heap.get_mut_ref::<V>(addr)) )
}
pub fn values_mut<'b: 'a>(&'b self,) -> impl Iterator<Item=&'a mut V> + 'b {
let heap: &'a Heap = self.heap;
let table: &'b [KeyValue] = &self.table;
self.occupied
.iter()
.cloned()
.map(move |bucket: usize| table[bucket].value_addr)
.map(move |addr: u32| heap.get_mut_ref::<V>(addr))
pub fn iter<'b: 'a>(&'b self) -> impl Iterator<Item = (&'a [u8], u32)> + 'b {
self.occupied.iter().cloned().map(move |bucket: usize| {
let kv = self.table[bucket];
self.get_key_value(kv.key_value_addr)
})
}
pub fn get_or_create<S: AsRef<[u8]>>(&mut self, key: S) -> &mut V {
let entry = self.lookup(key.as_ref());
match entry {
Entry::Occupied(addr) => {
self.heap.get_mut_ref(addr)
}
Entry::Vacant(bucket) => {
let (addr, val): (u32, &mut V) = self.heap.allocate_object();
self.set_bucket(key.as_ref(), bucket, addr);
val
}
}
}
pub fn lookup<S: AsRef<[u8]>>(&self, key: S) -> Entry {
pub fn get_or_create<S: AsRef<[u8]>, V: HeapAllocable>(&mut self, key: S) -> &mut V {
let key_bytes: &[u8] = key.as_ref();
let mut bucket = self.bucket(key_bytes);
let hash = murmurhash2::murmurhash2(key.as_ref());
let mut probe = self.probe(hash);
loop {
let bucket = probe.next_probe();
let kv: KeyValue = self.table[bucket];
if kv.is_empty() {
return Entry::Vacant(bucket);
let key_bytes_ref = self.heap.allocate_and_set(key_bytes);
let (addr, val): (u32, &mut V) = self.heap.allocate_object();
assert_eq!(addr, key_bytes_ref.addr() + 2 + key_bytes.len() as u32);
self.set_bucket(hash, key_bytes_ref, bucket);
return val;
} else if kv.hash == hash {
let (stored_key, expull_addr): (&[u8], u32) = self.get_key_value(kv.key_value_addr);
if stored_key == key_bytes {
return self.heap.get_mut_ref(expull_addr);
}
}
if self.get_key(kv.key) == key_bytes {
return Entry::Occupied(kv.value_addr);
}
bucket = (bucket + 1) & self.mask;
}
}
}
#[cfg(test)]
mod tests {
use super::*;
use super::super::heap::{Heap, HeapAllocable};
use super::djb2;
use super::murmurhash2::murmurhash2;
use test::Bencher;
use std::collections::hash_map::DefaultHasher;
use std::hash::Hasher;
use std::collections::HashSet;
use super::split_memory;
struct TestValue {
val: u32,
@@ -180,17 +215,21 @@ mod tests {
}
}
#[test]
fn test_hashmap_size() {
assert_eq!(split_memory(100_000), (67232, 12));
assert_eq!(split_memory(1_000_000), (737856, 15));
assert_eq!(split_memory(10_000_000), (7902848, 18));
}
#[test]
fn test_hash_map() {
let heap = Heap::with_capacity(2_000_000);
let mut hash_map: HashMap<TestValue> = HashMap::new(18, &heap);
let mut hash_map: HashMap = HashMap::new(18, &heap);
{
{
let v: &mut TestValue = hash_map.get_or_create("abc");
assert_eq!(v.val, 0u32);
v.val = 3u32;
}
}
{
let v: &mut TestValue = hash_map.get_or_create("abcd");
@@ -205,29 +244,55 @@ mod tests {
let v: &mut TestValue = hash_map.get_or_create("abcd");
assert_eq!(v.val, 4u32);
}
let mut iter_values = hash_map.values_mut();
assert_eq!(iter_values.next().unwrap().val, 3u32);
assert_eq!(iter_values.next().unwrap().val, 4u32);
assert!(!iter_values.next().is_some());
let mut iter_values = hash_map.iter();
{
let (_, addr) = iter_values.next().unwrap();
let val: &TestValue = heap.get_ref(addr);
assert_eq!(val.val, 3u32);
}
{
let (_, addr) = iter_values.next().unwrap();
let val: &TestValue = heap.get_ref(addr);
assert_eq!(val.val, 4u32);
}
assert!(iter_values.next().is_none());
}
#[test]
fn test_murmur() {
let s1 = "abcdef";
let s2 = "abcdeg";
for i in 0..5 {
assert_eq!(
murmurhash2(&s1[i..5].as_bytes()),
murmurhash2(&s2[i..5].as_bytes())
);
}
}
#[test]
fn test_murmur_collisions() {
let mut set: HashSet<u32> = HashSet::default();
for i in 0..10_000 {
let s = format!("hash{}", i);
let hash = murmurhash2(s.as_bytes());
set.insert(hash);
}
assert_eq!(set.len(), 10_000);
}
#[bench]
fn bench_djb2(bench: &mut Bencher) {
let v = String::from("abwer");
bench.iter(|| {
djb2(v.as_bytes())
});
}
#[bench]
fn bench_siphasher(bench: &mut Bencher) {
let v = String::from("abwer");
bench.iter(|| {
let mut h = DefaultHasher::new();
h.write(v.as_bytes());
h.finish()
fn bench_murmurhash_2(b: &mut Bencher) {
let keys: Vec<&'static str> =
vec!["wer qwe qwe qwe ", "werbq weqweqwe2 ", "weraq weqweqwe3 "];
b.iter(|| {
keys.iter()
.map(|&s| s.as_bytes())
.map(murmurhash2::murmurhash2)
.map(|h| h as u64)
.last()
.unwrap()
});
}
}

183
src/datastruct/stacker/heap.rs
View File

@@ -1,12 +1,29 @@
use std::cell::UnsafeCell;
use std::mem;
use std::ptr;
use byteorder::{ByteOrder, NativeEndian};
/// `BytesRef` refers to a slice in tantivy's custom `Heap`.
///
/// The slice will encode the length of the `&[u8]` slice
/// on 16-bits, and then the data is encoded.
#[derive(Copy, Clone)]
pub struct BytesRef {
pub start: u32,
pub stop: u32,
pub struct BytesRef(u32);
impl BytesRef {
pub fn is_null(&self) -> bool {
self.0 == u32::max_value()
}
pub fn addr(&self) -> u32 {
self.0
}
}
impl Default for BytesRef {
fn default() -> BytesRef {
BytesRef(u32::max_value())
}
}
/// Object that can be allocated in tantivy's custom `Heap`.
@@ -19,20 +36,19 @@ pub struct Heap {
inner: UnsafeCell<InnerHeap>,
}
#[cfg_attr(feature = "cargo-clippy", allow(mut_from_ref))]
impl Heap {
/// Creates a new heap with a given capacity
pub fn with_capacity(num_bytes: usize) -> Heap {
Heap {
inner: UnsafeCell::new(
InnerHeap::with_capacity(num_bytes)
),
inner: UnsafeCell::new(InnerHeap::with_capacity(num_bytes)),
}
}
fn inner(&self,) -> &mut InnerHeap {
unsafe { &mut *self.inner.get() }
fn inner(&self) -> &mut InnerHeap {
unsafe { &mut *self.inner.get() }
}
/// Clears the heap. All the underlying data is lost.
///
/// This heap does not support deallocation.
@@ -40,20 +56,9 @@ impl Heap {
pub fn clear(&self) {
self.inner().clear();
}
/// Return the heap capacity.
pub fn capacity(&self,) -> u32 {
self.inner().capacity()
}
/// Return the amount of memory that has been allocated so far.
pub fn len(&self,) -> u32 {
self.inner().len()
}
/// Return amount of free space, in bytes.
pub fn num_free_bytes(&self,) -> u32 {
pub fn num_free_bytes(&self) -> u32 {
self.inner().num_free_bytes()
}
@@ -62,37 +67,42 @@ impl Heap {
pub fn allocate_space(&self, num_bytes: usize) -> u32 {
self.inner().allocate_space(num_bytes)
}
/// Allocate an object in the heap
pub fn allocate_object<V: HeapAllocable>(&self,) -> (u32, &mut V) {
pub fn allocate_object<V: HeapAllocable>(&self) -> (u32, &mut V) {
let addr = self.inner().allocate_space(mem::size_of::<V>());
let v: V = V::with_addr(addr);
self.inner().set(addr, &v);
(addr, self.inner().get_mut_ref(addr))
}
/// Stores a `&[u8]` in the heap and returns the destination BytesRef.
pub fn allocate_and_set(&self, data: &[u8]) -> BytesRef {
self.inner().allocate_and_set(data)
}
/// Fetches the `&[u8]` stored on the slice defined by the `BytesRef`
/// given as argumetn
pub fn get_slice(&self, bytes_ref: BytesRef) -> &[u8] {
self.inner().get_slice(bytes_ref.start, bytes_ref.stop)
self.inner().get_slice(bytes_ref)
}
/// Stores an item's data in the heap, at the given `address`.
pub fn set<Item>(&self, addr: u32, val: &Item) {
self.inner().set(addr, val);
}
/// Returns a mutable reference for an object at a given Item.
pub fn get_mut_ref<Item>(&self, addr: u32) -> &mut Item {
self.inner().get_mut_ref(addr)
}
}
/// Returns a mutable reference to an `Item` at a given `addr`.
#[cfg(test)]
pub fn get_ref<Item>(&self, addr: u32) -> &mut Item {
self.get_mut_ref(addr)
}
}
struct InnerHeap {
buffer: Vec<u8>,
@@ -101,24 +111,11 @@ struct InnerHeap {
next_heap: Option<Box<InnerHeap>>,
}
/// initializing a long Vec<u8> is crazy slow in
/// debug mode.
/// We use this unsafe trick to make unit test
/// way faster.
fn allocate_fast(num_bytes: usize) -> Vec<u8> {
let mut buffer = Vec::with_capacity(num_bytes);
unsafe {
buffer.set_len(num_bytes);
}
buffer
}
impl InnerHeap {
pub fn with_capacity(num_bytes: usize) -> InnerHeap {
let buffer: Vec<u8> = allocate_fast(num_bytes);
let buffer: Vec<u8> = vec![0u8; num_bytes];
InnerHeap {
buffer: buffer,
buffer,
buffer_len: num_bytes as u32,
next_heap: None,
used: 0u32,
@@ -130,23 +127,14 @@ impl InnerHeap {
self.next_heap = None;
}
pub fn capacity(&self,) -> u32 {
self.buffer.len() as u32
}
pub fn len(&self,) -> u32 {
self.used
}
// Returns the number of free bytes. If the buffer
// has reached it's capacity and overflowed to another buffer, return 0.
pub fn num_free_bytes(&self,) -> u32 {
pub fn num_free_bytes(&self) -> u32 {
if self.next_heap.is_some() {
0u32
}
else {
} else {
self.buffer_len - self.used
}
}
}
pub fn allocate_space(&mut self, num_bytes: usize) -> u32 {
@@ -154,74 +142,85 @@ impl InnerHeap {
self.used += num_bytes as u32;
if self.used <= self.buffer_len {
addr
}
else {
} else {
if self.next_heap.is_none() {
warn!("Exceeded heap size. The margin was apparently unsufficient. The segment will be committed right after indexing this very last document.");
info!(
r#"Exceeded heap size. The segment will be committed right
after indexing this document."#,
);
self.next_heap = Some(Box::new(InnerHeap::with_capacity(self.buffer_len as usize)));
}
self.next_heap.as_mut().unwrap().allocate_space(num_bytes) + self.buffer_len
}
}
fn get_slice(&self, start: u32, stop: u32) -> &[u8] {
fn get_slice(&self, bytes_ref: BytesRef) -> &[u8] {
let start = bytes_ref.0;
if start >= self.buffer_len {
self.next_heap.as_ref().unwrap().get_slice(start - self.buffer_len, stop - self.buffer_len)
}
else {
&self.buffer[start as usize..stop as usize]
self.next_heap
.as_ref()
.unwrap()
.get_slice(BytesRef(start - self.buffer_len))
} else {
let start = start as usize;
let len = NativeEndian::read_u16(&self.buffer[start..start + 2]) as usize;
&self.buffer[start + 2..start + 2 + len]
}
}
fn get_mut_slice(&mut self, start: u32, stop: u32) -> &mut [u8] {
if start >= self.buffer_len {
self.next_heap.as_mut().unwrap().get_mut_slice(start - self.buffer_len, stop - self.buffer_len)
}
else {
self.next_heap
.as_mut()
.unwrap()
.get_mut_slice(start - self.buffer_len, stop - self.buffer_len)
} else {
&mut self.buffer[start as usize..stop as usize]
}
}
fn allocate_and_set(&mut self, data: &[u8]) -> BytesRef {
let start = self.allocate_space(data.len());
let stop = start + data.len() as u32;
self.get_mut_slice(start, stop).clone_from_slice(data);
BytesRef {
start: start as u32,
stop: stop as u32,
}
assert!(data.len() < u16::max_value() as usize);
let total_len = 2 + data.len();
let start = self.allocate_space(total_len);
let total_buff = self.get_mut_slice(start, start + total_len as u32);
NativeEndian::write_u16(&mut total_buff[0..2], data.len() as u16);
total_buff[2..].clone_from_slice(data);
BytesRef(start)
}
fn get_mut(&mut self, addr: u32) -> *mut u8 {
if addr >= self.buffer_len {
self.next_heap.as_mut().unwrap().get_mut(addr - self.buffer_len)
}
else {
self.next_heap
.as_mut()
.unwrap()
.get_mut(addr - self.buffer_len)
} else {
let addr_isize = addr as isize;
unsafe { self.buffer.as_mut_ptr().offset(addr_isize) }
}
}
fn get_mut_ref<Item>(&mut self, addr: u32) -> &mut Item {
if addr >= self.buffer_len {
self.next_heap.as_mut().unwrap().get_mut_ref(addr - self.buffer_len)
}
else {
self.next_heap
.as_mut()
.unwrap()
.get_mut_ref(addr - self.buffer_len)
} else {
let v_ptr_u8 = self.get_mut(addr) as *mut u8;
let v_ptr = v_ptr_u8 as *mut Item;
unsafe { &mut *v_ptr }
}
}
fn set<Item>(&mut self, addr: u32, val: &Item) {
pub fn set<Item>(&mut self, addr: u32, val: &Item) {
if addr >= self.buffer_len {
self.next_heap.as_mut().unwrap().set(addr - self.buffer_len, val);
}
else {
self.next_heap
.as_mut()
.unwrap()
.set(addr - self.buffer_len, val);
} else {
let v_ptr: *const Item = val as *const Item;
let v_ptr_u8: *const u8 = v_ptr as *const u8;
debug_assert!(addr + mem::size_of::<Item>() as u32 <= self.used);
@@ -231,4 +230,4 @@ impl InnerHeap {
}
}
}
}
}

43
src/datastruct/stacker/mod.rs
View File

@@ -1,46 +1,43 @@
mod hashmap;
pub(crate) mod hashmap;
mod heap;
mod expull;
pub use self::heap::{Heap, HeapAllocable};
pub use self::expull::ExpUnrolledLinkedList;
pub use self::hashmap::{HashMap, Entry};
pub use self::hashmap::HashMap;
#[test]
fn test_unrolled_linked_list() {
use std::collections;
let heap = Heap::with_capacity(30_000_000);
{
heap.clear();
let mut ks: Vec<usize> = (1..5).map(|k| k * 100).collect();
ks.push(2);
ks.push(3);
for k in (1..5).map(|k| k * 100) {
let mut hashmap: HashMap<ExpUnrolledLinkedList> = HashMap::new(10, &heap);
for k in (1..5).map(|k| k * 100) {
let mut hashmap: HashMap = HashMap::new(10, &heap);
for j in 0..k {
for i in 0..500 {
let mut list = hashmap.get_or_create(i.to_string());
list.push(i*j, &heap);
let v: &mut ExpUnrolledLinkedList = hashmap.get_or_create(i.to_string());
v.push(i * j, &heap);
}
}
let mut map_addr: collections::HashMap<Vec<u8>, u32> = collections::HashMap::new();
for (key, addr) in hashmap.iter() {
map_addr.insert(Vec::from(key), addr);
}
for i in 0..500 {
match hashmap.lookup(i.to_string()) {
Entry::Occupied(addr) => {
let v: &mut ExpUnrolledLinkedList = heap.get_mut_ref(addr);
let mut it = v.iter(addr, &heap);
for j in 0..k {
assert_eq!(it.next().unwrap(), i*j);
}
assert!(!it.next().is_some());
}
_ => {
panic!("should never happen");
}
let key: String = i.to_string();
let addr: u32 = *map_addr.get(key.as_bytes()).unwrap();
let exp_pull: &ExpUnrolledLinkedList = heap.get_ref(addr);
let mut it = exp_pull.iter(addr, &heap);
for j in 0..k {
assert_eq!(it.next().unwrap(), i * j);
}
assert!(!it.next().is_some());
}
}
}
}
}

57
src/directory/directory.rs
View File

@@ -1,56 +1,56 @@
use std::marker::Send;
use std::fmt;
use std::path::Path;
use directory::error::{FileError, OpenWriteError};
use directory::error::{DeleteError, OpenReadError, OpenWriteError};
use directory::{ReadOnlySource, WritePtr};
use std::result;
use std::io;
use std::marker::Sync;
/// Write-once read many (WORM) abstraction for where tantivy's index should be stored.
/// Write-once read many (WORM) abstraction for where
/// tantivy's data should be stored.
///
/// There are currently two implementations of `Directory`
///
///
/// - The [`MMapDirectory`](struct.MmapDirectory.html), this
/// should be your default choice.
/// - The [`RAMDirectory`](struct.RAMDirectory.html), which
/// should be your default choice.
/// - The [`RAMDirectory`](struct.RAMDirectory.html), which
/// should be used mostly for tests.
///
///
pub trait Directory: fmt::Debug + Send + Sync + 'static {
/// Opens a virtual file for read.
///
///
/// Once a virtual file is open, its data may not
/// change.
///
/// Specifically, subsequent writes or flushes should
/// have no effect on the returned `ReadOnlySource` object.
fn open_read(&self, path: &Path) -> result::Result<ReadOnlySource, FileError>;
/// have no effect on the returned `ReadOnlySource` object.
fn open_read(&self, path: &Path) -> result::Result<ReadOnlySource, OpenReadError>;
/// Removes a file
///
/// Removing a file will not affect an eventual
/// existing ReadOnlySource pointing to it.
///
///
/// Removing a nonexistent file, yields a
/// `FileError::DoesNotExist`.
fn delete(&self, path: &Path) -> result::Result<(), FileError>;
/// `DeleteError::DoesNotExist`.
fn delete(&self, path: &Path) -> result::Result<(), DeleteError>;
/// Returns true iff the file exists
fn exists(&self, path: &Path) -> bool;
/// Opens a writer for the *virtual file* associated with
/// Opens a writer for the *virtual file* associated with
/// a Path.
///
/// Right after this call, the file should be created
/// and any subsequent call to `open_read` for the
/// and any subsequent call to `open_read` for the
/// same path should return a `ReadOnlySource`.
///
///
/// Write operations may be aggressively buffered.
/// The client of this trait is responsible for calling flush
/// to ensure that subsequent `read` operations
/// to ensure that subsequent `read` operations
/// will take into account preceding `write` operations.
///
///
/// Flush operation should also be persistent.
///
/// The user shall not rely on `Drop` triggering `flush`.
@@ -59,18 +59,21 @@ pub trait Directory: fmt::Debug + Send + Sync + 'static {
///
/// The file may not previously exist.
fn open_write(&mut self, path: &Path) -> Result<WritePtr, OpenWriteError>;
/// Reads the full content file that has been written using
/// atomic_write.
///
/// This should only be used for small files.
fn atomic_read(&self, path: &Path) -> Result<Vec<u8>, OpenReadError>;
/// Atomically replace the content of a file with data.
///
///
/// This calls ensure that reads can never *observe*
/// a partially written file.
///
///
/// The file may or may not previously exist.
fn atomic_write(&mut self, path: &Path, data: &[u8]) -> io::Result<()>;
/// Clones the directory and boxes the clone
/// Clones the directory and boxes the clone
fn box_clone(&self) -> Box<Directory>;
}

Some files were not shown because too many files have changed in this diff Show More