[MOD-8198] Introduce INT8 distance functions #560

meiravgri · 2024-12-03T21:56:17Z

This PR introduces naive and optimized distance calculation functions for vectors of INT8 type elements, for all distance metrics.
Since INT8 vectors store integral values, we can't normalize these vectors in place.
Because of this, unlike other types, INT8 requires a dedicated implementation for cosine distance.

⚠️ The cosine implementation assumes the norm is stored at the end of the vector in a float number.

Optimized distance functions

Requirments

Optimization is available for CPUs and compilers supporting all of the following flags:

AVX512_F
AVX512_BW
AVX512_VL
AVX512_VNNI

Supported dimensions

Optimization is available for vector dimensions larger than 32.
Vectors with smaller dimensions will be calculated using naive implementation.

alignment

IP and L2

The vectors memory is aligned to memory addresses divisible by 32 * sizeof(int8) since In each step we load 32-int8 elements

Cosine

We skip aligning the vector's memory address for int8 vectors with cosine distance due to the extra float stored at the end for the norm.
This changes the alignment requirement to (dim + sizeof(float)) % 32 for the vector's address to be aligned to 256 bits.
Since vectors that meet this condition necessarily have a residual (since dim % 32 ≠ 0), subsequent loads will be offset during distance calculation.
To avoid adding complexity to the distance function, we do not align the vectors in this case,
assuming the performance impact is negligible.

Changes in tests

changed unit/test_utils.* files name to unit/unit_test_utils.*
added general tests_utils file to tests/utils

codecov · 2024-12-03T22:29:42Z

Codecov Report

Attention: Patch coverage is 95.91837% with 6 lines in your changes missing coverage. Please review.

Project coverage is 96.91%. Comparing base (35ddc2e) to head (c4439f3).
Report is 2 commits behind head on meiravg_feature_int_uint_8.

❗ Current head c4439f3 differs from pull request most recent head e526d02

Please upload reports for the commit e526d02 to get more accurate results.

Files with missing lines	Patch %	Lines
src/VecSim/spaces/spaces.cpp	33.33%	6 Missing ⚠️

Additional details and impacted files

@@                      Coverage Diff                       @@
##           meiravg_feature_int_uint_8     #560      +/-   ##
==============================================================
- Coverage                       96.97%   96.91%   -0.06%     
==============================================================
  Files                             100      103       +3     
  Lines                            5287     5444     +157     
==============================================================
+ Hits                             5127     5276     +149     
- Misses                            160      168       +8

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

add cosine to spaces fix typos in calculator

add spaces unit tests for int8 L2 add compilation flags introduce tests/utils for general utils

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

change create_int8_vec to populate_int8_vec add compute norm

minimal dim = 32

alonre24

💪
Few small comments/questions

src/VecSim/spaces/IP/IP.cpp

alonre24 · 2024-12-15T07:53:48Z

src/VecSim/spaces/IP_space.cpp

+    return ret_dist_func;
+}
+
+dist_func_t<float> Cosine_INT8_GetDistFunc(size_t dim, unsigned char *alignment,


This function has the exact same logic as the above except for the call to Choose_INT8_Cosine_implementation_AVX512F_BW_VL_VNNI eight. Consider doing a consolidation for the common logic

The only common code is the alignment intizliation, which we duplicate in all the functions in this file, how is this function different?

btw note to self: check coverage

coverage: added cosine to small dimension spaces unit tests

src/VecSim/spaces/IP/IP_AVX512F_BW_VL_VNNI_INT8.h

src/VecSim/spaces/IP/IP.cpp

src/VecSim/spaces/L2/L2.cpp

src/VecSim/spaces/spaces.cpp

tests/unit/test_spaces.cpp

src/VecSim/spaces/L2/L2_AVX512F_BW_VL_VNNI_INT8.h

src/VecSim/spaces/IP_space.cpp

GuyAv46 · 2024-12-15T14:22:09Z

src/VecSim/spaces/IP/IP_AVX512F_BW_VL_VNNI_INT8.h

+    // Multiply groups of 2 adjacent pairs of signed 16-bit integers in `a` with corresponding
+    // 16-bit integers in `b`, producing 2 intermediate signed 32-bit results. Sum these 2 results
+    // with the corresponding 32-bit integer in src, and store the packed 32-bit results in dst.
+    sum = _mm512_dpwssd_epi32(sum, va, vb);


Why not use _mm512_dpbusd_epi32, avoiding loading only 32 elements and converting them to int16?

nice catch!
Unfortunately, _mm512_dpbusd_epi32 expects a to be unsigned 8-bit integers vecor, and b to be signed 8-bit vector

Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in a with corresponding signed 8-bit integers in b

So although indeed faster, it gives the wrong results

Benchmark Time CPU Iterations UserCounters... -------------------------------------------------------------------------------------------- IP_dpwssd/Dim:512 5.41 ns 5.41 ns 128425185 dist=-1.024k IP_dpbusd/Dim:512 3.26 ns 3.26 ns 218138693 dist=130.048k IP_sanity_bm/Dim:512/iterations:1 490 ns 360 ns 1 dist=-1.024k

Bummer... So there is no function for multiplying int8int8 or uint8uint8? Only an unsigned by a signed?

nope
we might want to consider using _mm512_unpackhi_epi8 and _mm512_unpacklo_epi8 for uint8_t

src/VecSim/spaces/IP/IP_AVX512F_BW_VL_VNNI_INT8.h

src/VecSim/spaces/IP/IP.cpp

src/VecSim/spaces/IP/IP_AVX512F_BW_VL_VNNI_INT8.h

align to vector size ncluding the norm in cosine dist unit test cover small dim in cosine chooser

GuyAv46

🤯

* 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 (cherry picked from commit afb5250)

github-actions · 2024-12-17T05:47:25Z

Successfully created backport PR for 8.0:

[8.0] [MOD-8198] Introduce INT8 distance functions #570

* [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

* [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)

[MOD-8198] Introduce INT8 (#560) (#571) * [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) Co-authored-by: meiravgri <109056284+meiravgri@users.noreply.github.com>

meiravgri added the bm-spaces label Dec 9, 2024

meiravgri changed the title ~~naive implementation of L2~~ [MOD-8198] Introduce INT8 distance functions Dec 10, 2024

meiravgri requested a review from GuyAv46 December 10, 2024 10:50

meiravgri added 17 commits December 10, 2024 16:09

naive implementation of L2

2f34c15

update

c641d23

implment naive disatnce for int8

1c5eb90

add cosine to spaces fix typos in calculator

imp choose L2 int8 with 256bit loop

fa8e9ff

add spaces unit tests for int8 L2 add compilation flags introduce tests/utils for general utils

imp space bm for int8

a7a556f

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

43064e8

rename unit/test_utuils to unit_test_utils

fb9f1cc

seed create vec

602f8e9

format

cde5e2d

implmenet IP + unit test

cdb4d7f

ip bm

5f01890

format

2dce6f0

implement cosine in ip API

3d3b375

change create_int8_vec to populate_int8_vec add compute norm

use mask sub instead of msk load

6f211b3

loop size = 512

6ac65a3

minimal dim = 32

add int8 to bm

0d07c5d

reanme to simd64

3586a76

meiravgri force-pushed the meiravg_int8_dist_func branch from d427c81 to 3586a76 Compare December 10, 2024 16:10

meiravgri changed the base branch from main to meiravg_feature_int_uint_8 December 10, 2024 16:13

convert to int before multiplication

adbc4d7

meiravgri requested a review from alonre24 December 12, 2024 12:10

alonre24 reviewed Dec 15, 2024

View reviewed changes

GuyAv46 reviewed Dec 15, 2024

View reviewed changes

meiravgri added 3 commits December 16, 2024 13:00

review comments:

cb2c887

align to vector size ncluding the norm in cosine dist unit test cover small dim in cosine chooser

use sizeof(float)instead of 4

880dd33

remove int conversion in test_utils::compute_norm

b79777f

meiravgri requested review from GuyAv46 and alonre24 December 16, 2024 13:11

meiravgri added 4 commits December 16, 2024 13:53

REVERT!!! malicious test to see if we get to the code

ab159bc

assert dummt

397ac3f

fix alignemnt test

f9b7b87

remove assert

c4439f3

meiravgri added the backport 8.0 label Dec 16, 2024

GuyAv46 approved these changes Dec 16, 2024

View reviewed changes

alonre24 approved these changes Dec 16, 2024

View reviewed changes

remove cosine alignment

e526d02

meiravgri merged commit afb5250 into meiravg_feature_int_uint_8 Dec 17, 2024
11 checks passed

meiravgri deleted the meiravg_int8_dist_func branch December 17, 2024 05:47

github-actions bot mentioned this pull request Dec 17, 2024

[8.0] [MOD-8198] Introduce INT8 distance functions #570

Closed

[MOD-8198] Introduce INT8 distance functions #560

[MOD-8198] Introduce INT8 distance functions #560

Uh oh!

Conversation

meiravgri commented Dec 3, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Optimized distance functions

Requirments

Supported dimensions

alignment

IP and L2

Cosine

Changes in tests

Uh oh!

codecov bot commented Dec 3, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

alonre24 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

alonre24 Dec 15, 2024

Choose a reason for hiding this comment

Uh oh!

meiravgri Dec 15, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

meiravgri Dec 15, 2024

Choose a reason for hiding this comment

Uh oh!

meiravgri Dec 16, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

GuyAv46 Dec 15, 2024

Choose a reason for hiding this comment

Uh oh!

meiravgri Dec 16, 2024

Choose a reason for hiding this comment

Uh oh!

GuyAv46 Dec 16, 2024

Choose a reason for hiding this comment

Uh oh!

meiravgri Dec 16, 2024

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

GuyAv46 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

github-actions bot commented Dec 17, 2024

Uh oh!

Uh oh!

meiravgri commented Dec 3, 2024 •

edited

Loading

codecov bot commented Dec 3, 2024 •

edited

Loading

meiravgri Dec 15, 2024 •

edited

Loading

meiravgri Dec 16, 2024 •

edited

Loading