[8.0] [MOD-8198] Introduce INT8 (#560) #577
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
* [MOD-8198] Introduce INT8 distance functions (#560) * naive implementation of L2 * update * implment naive disatnce for int8 add cosine to spaces fix typos in calculator * imp choose L2 int8 with 256bit loop add spaces unit tests for int8 L2 add compilation flags introduce tests/utils for general utils * imp space bm for int8 change INITIALIZE_BENCHMARKS_SET to INITIALIZE_BENCHMARKS_SET_L2_IP introduce INITIALIZE_BENCHMARKS_SET_COSINE fix typos in Choose_INT8_L2_implementation_AVX512F_BW_VL_VNNI name * fix INITIALIZE_BENCHMARKS_SET_L2_IP and add include to F_BW_VL_VNNI * rename unit/test_utuils to unit_test_utils * seed create vec * format * implmenet IP + unit test * ip bm * format * implement cosine in ip API change create_int8_vec to populate_int8_vec add compute norm * use mask sub instead of msk load * loop size = 512 minimal dim = 32 * add int8 to bm * reanme to simd64 * convert to int before multiplication * review comments: align to vector size ncluding the norm in cosine dist unit test cover small dim in cosine chooser * use sizeof(float)instead of 4 * remove int conversion in test_utils::compute_norm * REVERT!!! malicious test to see if we get to the code * assert dummt * fix alignemnt test * remove assert * remove cosine alignment * Override missing intrinsincs in gcc <11 (#572) * override _mm256_loadu_epi8 with mm256_maskz_loadu_epi8 if gcc < 11 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95483 * fix * disable flow temp * add comment * [MOD-8200] [MOD-8202] INT8 index (#566) * naive implementation of L2 * update * implment naive disatnce for int8 add cosine to spaces fix typos in calculator * imp choose L2 int8 with 256bit loop add spaces unit tests for int8 L2 add compilation flags introduce tests/utils for general utils * imp space bm for int8 change INITIALIZE_BENCHMARKS_SET to INITIALIZE_BENCHMARKS_SET_L2_IP introduce INITIALIZE_BENCHMARKS_SET_COSINE fix typos in Choose_INT8_L2_implementation_AVX512F_BW_VL_VNNI name * fix INITIALIZE_BENCHMARKS_SET_L2_IP and add include to F_BW_VL_VNNI * rename unit/test_utuils to unit_test_utils * seed create vec * format * implmenet IP + unit test * ip bm * format * implement cosine in ip API change create_int8_vec to populate_int8_vec add compute norm * use mask sub instead of msk load * loop size = 512 minimal dim = 32 * add int8 to bm * reanme to simd64 * convert to int before multiplication * introduce IntegralType_ComputeNorm * move preprocessor logic to choose if cosine preprocessor is needed to CreateIndexComponents: pass bool is_normalized get distnce function according to original metric get pp according to is_normalized && metric == VecSimMetric_Cosine, and remove this logic from the indexes factories. add dataSize member to AbstractIndexInitParams add VecSimType_INT8 type introduce VecSimParams_GetDataSize: returns datasize introduce and implement GetNormalizeFunc<int8_t> thtat returns int8_normalizeVector int8_normalizeVector computes the norm and stores it at the emd of argument vector. * add int8 tests * fix include unint_test_utils * add int 8 to index factories remove normalize_func from VecSimIndexAbstract members tests: int8 unit test create int8 indexes unit_test_utils: CalcIndexDataSize: casts VecSimIndex * to VecSimIndexAbstract<dist_t, data_t> * and calls VecSimIndexAbstract<dist_t, data_t>::getDataSize() cast_to_tiered_index<data_t, dist_t>: takes VecSimIndex * ans casts to TieredHNSWIndex<data_t, dist_t> * * add EstimateInitialSize for int8 to indexes factories 2 new function to test_utils:: CreateTieredParams CreateNewTieredHNSWIndex add test_initial_size_estimation to CommonTypeMetricTests use CommonTypeMetricTieredTests for tiered tests * add int8 unit tests add int8 to * VecSimDebug_GetElementNeighborsInHNSWGraph * VecSim_Normalize *HNSW NewIndex from file * remove duplicated GetDistFunc<int8_t, float> move ASSERT_DEBUG_DEATH of CalcIndexDataSize to a separate test * remove assert test, the statement is excuted and causes crash * imporve normalize test * rename test_utils::compute_norm -> test_utils::integral_compute_norm remove test_normalize.cpp file * use stack allocation instead of heap allocation in tests * fix float comparison in test_serialization avoid evaluating statement in typeid to avoid clang warnig * renae CalcIndexDataSize -> CalcVectorDataSize move components tests from test_common to test_components * add comment to INSTANTIATE_TEST_SUITE_P * [MOD-8206] INT8 flow tests (#573) * test_hnsw.py intiital * int8 hnsw tests * general tests class * flow_bruteforce.py: introduce GeneralTest call from TestINT8 common.py: introduce create_flat_index create_add_vectors move fp32_expand_and_calc_cosine_dist to common.py * tiered flow tests: * add optional create_data_func to IndexCtx, use for special datatypes *inntroduce test_create_int8 and test_search_insert_int8 create_int8_vectors expectes shape (tuple) * use query.flat * revert using flat (not helping in int8) fix float16 calling query.flat * revert changes in Data class in bf tests revert test_bf_float16_range_query change * fix merge (cherry picked from commit babfbe0)
meiravgri
approved these changes
Dec 24, 2024
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## 8.0 #577 +/- ##
==========================================
+ Coverage 96.98% 97.09% +0.10%
==========================================
Files 100 104 +4
Lines 5305 5503 +198
==========================================
+ Hits 5145 5343 +198
Misses 160 160 ☔ View full report in Codecov by Sentry. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Backport of #571 to
8.0.