Rename dist functions from SQ8 to SQ8-FP32 by dor-forer · Pull Request #888 · RedisAI/VectorSimilarity

dor-forer · 2026-01-12T13:22:23Z

Describe the changes in the pull request

Rename SQ8 distance functions and files to SQ8_FP32 to clarify that they compute asymmetric distance between SQ8 (quantized storage) and FP32 (query) vectors.

This distinguishes them from SQ8_SQ8 functions, which compute the symmetric distance between two SQ8 vectors.

Changes include:

Rename distance functions: SQ8_* → SQ8_FP32_* (InnerProduct, Cosine, L2Sqr)
Rename implementation selectors: Choose_SQ8_* → Choose_SQ8_FP32_*
Rename header files: *_SQ8.h → *_SQ8_FP32.h for IP and L2 spaces
Rename benchmark file: bm_spaces_sq8.cpp → bm_spaces_sq8_fp32.cpp

Which issues this PR fixes

MOD-13407

Main objects this PR modified

...
...

Mark if applicable

This PR introduces API changes
This PR introduces serialization changes

Copilot

Pull request overview

This pull request renames SQ8 distance functions and files to SQ8_FP32 to distinguish asymmetric distance computation (between SQ8 storage and FP32 query) from symmetric SQ8_SQ8 distance computation (between two SQ8 vectors).

Changes:

Renamed distance functions from SQ8_* to SQ8_FP32_* for InnerProduct, Cosine, and L2Sqr operations
Updated implementation selectors from Choose_SQ8_* to Choose_SQ8_FP32_* across all SIMD architectures
Corrected parameter order in function signatures to reflect SQ8 storage as first parameter and FP32 query as second parameter
Renamed benchmark files and test utilities to match new naming convention

Reviewed changes

Copilot reviewed 42 out of 42 changed files in this pull request and generated 11 comments.

Show a summary per file

File	Description
tests/utils/tests_utils.h	Renamed helper functions and corrected parameter order for SQ8-FP32 distance calculations
tests/unit/test_spaces.cpp	Updated test names and function calls; contains critical bugs with SQ8_SQ8 test names incorrectly renamed
tests/benchmark/spaces_benchmarks/bm_spaces_sq8_fp32.cpp	Renamed benchmark class and updated function calls
tests/benchmark/benchmarks.sh	Updated benchmark script references from spaces_sq8 to spaces_sq8_fp32
tests/benchmark/CMakeLists.txt	Updated DATA_TYPE list to use sq8_fp32 instead of sq8
src/VecSim/spaces/spaces.cpp	Updated GetDistFunc to call renamed SQ8_FP32 functions
src/VecSim/spaces/functions/*.h	Renamed function declarations for all SIMD architectures (SVE, SVE2, SSE4, NEON, AVX2, AVX2_FMA, AVX512)
src/VecSim/spaces/functions/*.cpp	Updated function implementations and choosers across all SIMD architectures
src/VecSim/spaces/L2_space.h	Renamed L2_SQ8_GetDistFunc to L2_SQ8_FP32_GetDistFunc with updated documentation
src/VecSim/spaces/L2_space.cpp	Updated function implementation and all SIMD path selections
src/VecSim/spaces/L2/L2_*.h	Renamed L2 template functions and corrected parameter order/comments across all SIMD implementations
src/VecSim/spaces/L2/L2.h	Updated function signature and comments
src/VecSim/spaces/L2/L2.cpp	Renamed implementation function and corrected parameter order
src/VecSim/spaces/IP_space.h	Renamed IP and Cosine GetDistFunc to SQ8_FP32 variants with updated documentation
src/VecSim/spaces/IP_space.cpp	Updated function implementations and all SIMD path selections
src/VecSim/spaces/IP/IP_*.h	Renamed IP template functions and corrected parameter order/comments across all SIMD implementations
src/VecSim/spaces/IP/IP.h	Updated function signatures and comments
src/VecSim/spaces/IP/IP.cpp	Renamed implementation function and corrected parameter order

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

tests/unit/test_spaces.cpp

Copilot · 2026-01-12T13:26:21Z

tests/unit/test_spaces.cpp

 /* ======================== Tests SQ8_SQ8 ========================= */

-TEST_F(SpacesTest, SQ8_SQ8_ip_no_optimization_func_test) {
+TEST_F(SpacesTest, SQ8_SQ8_FP32_ip_no_optimization_func_test) {


The test name is incorrectly updated. This test is for SQ8-to-SQ8 symmetric distance (both vectors are SQ8 quantized), not for SQ8-FP32 asymmetric distance. The test uses SQ8_SQ8 functions, so the name should remain "SQ8_SQ8_ip_no_optimization_func_test".

he has a point

tests/unit/test_spaces.cpp

Copilot · 2026-01-12T13:26:22Z

tests/unit/test_spaces.cpp


 // Test with constant vector (all same values) for L2
-TEST(SQ8_SQ8_EdgeCases, L2ConstantVectorTest) {
+TEST(SQ8_SQ8_FP32_EdgeCases, L2ConstantVectorTest) {


The test suite name is incorrectly updated. It should remain "SQ8_SQ8_EdgeCases" as this test is for SQ8-to-SQ8 symmetric distance (both vectors are SQ8 quantized), not for SQ8-FP32 asymmetric distance. The test content uses SQ8_SQ8_L2Sqr which is a different function from SQ8_FP32_L2Sqr.

tests/unit/test_spaces.cpp

- Rename distance functions: SQ8_* -> SQ8_FP32_* for IP, Cosine, L2 - Update variable ordering to clarify pVect1=SQ8 storage, pVect2=FP32 query - Update test utilities and benchmark references - Rename Choose_SQ8_* -> Choose_SQ8_FP32_* implementation selectors

- Rename IP headers: IP_*_SQ8.h -> IP_*_SQ8_FP32.h - Rename L2 headers: L2_*_SQ8.h -> L2_*_SQ8_FP32.h - Rename benchmark file: bm_spaces_sq8.cpp -> bm_spaces_sq8_fp32.cpp

Copilot

Pull request overview

Copilot reviewed 42 out of 42 changed files in this pull request and generated no new comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

codecov · 2026-01-12T13:42:29Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.09%. Comparing base (8a568cb) to head (18c1a81).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #888      +/-   ##
==========================================
- Coverage   97.11%   97.09%   -0.03%     
==========================================
  Files         129      129              
  Lines        7493     7493              
==========================================
- Hits         7277     7275       -2     
- Misses        216      218       +2

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

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

meiravgri

noice

meiravgri · 2026-01-13T14:30:50Z

src/VecSim/spaces/IP/IP_AVX2_SQ8_FP32.h

 // Helper: compute Σ(q_i * y_i) for 8 elements (no dequantization)
-static inline void InnerProductStepSQ8(const float *&pVect1, const uint8_t *&pVect2,
+// pVect1 = SQ8 storage (quantized values), pVect2 = FP32 query
+static inline void InnerProductStepSQ8(const uint8_t *&pVect1, const float *&pVect2,


Suggested change

static inline void InnerProductStepSQ8(const uint8_t *&pVect1, const float *&pVect2,

static inline void InnerProductStepSQ8_FP32(const uint8_t *&pVect1, const float *&pVect2,

meiravgri · 2026-01-13T14:31:35Z

src/VecSim/spaces/IP/IP_AVX512F_BW_VL_VNNI_SQ8_FP32.h

-    __m512 v1 = _mm512_loadu_ps(pVec1);
-
+// pVec1 = SQ8 storage (quantized values), pVec2 = FP32 query
+static inline void SQ8_InnerProductStep(const uint8_t *&pVec1, const float *&pVec2, __m512 &sum) {


Suggested change

static inline void SQ8_InnerProductStep(const uint8_t *&pVec1, const float *&pVec2, __m512 &sum) {

static inline void SQ8_FP32_InnerProductStep(const uint8_t *&pVec1, const float *&pVec2, __m512 &sum) {

meiravgri · 2026-01-13T14:32:15Z

src/VecSim/spaces/IP/IP_NEON_SQ8_FP32.h

 // Helper: compute Σ(q_i * y_i) for 4 elements (no dequantization)
-static inline void InnerProductStepSQ8(const float *&pVect1, const uint8_t *&pVect2,
+// pVect1 = SQ8 storage (quantized values), pVect2 = FP32 query
+static inline void InnerProductStepSQ8(const uint8_t *&pVect1, const float *&pVect2,


Suggested change

static inline void InnerProductStepSQ8(const uint8_t *&pVect1, const float *&pVect2,

static inline void InnerProductStepSQ8_FP32(const uint8_t *&pVect1, const float *&pVect2,

meiravgri · 2026-01-13T14:33:04Z

src/VecSim/spaces/IP/IP_SSE4_SQ8_FP32.h

-    // Load 4 float elements from query
-    __m128 v1 = _mm_loadu_ps(pVect1);
+// pVect1 = SQ8 storage (quantized values), pVect2 = FP32 query
+static inline void InnerProductStepSQ8(const uint8_t *&pVect1, const float *&pVect2, __m128 &sum) {


Suggested change

static inline void InnerProductStepSQ8(const uint8_t *&pVect1, const float *&pVect2, __m128 &sum) {

static inline void InnerProductStepSQ8_FP32(const uint8_t *&pVect1, const float *&pVect2, __m128 &sum) {

meiravgri · 2026-01-13T14:33:18Z

src/VecSim/spaces/IP/IP_SVE_SQ8_FP32.h

 // Helper: compute Σ(q_i * y_i) for one SVE vector width (no dequantization)
-static inline void InnerProductStepSQ8(const float *pVect1, const uint8_t *pVect2, size_t &offset,
+// pVect1 = SQ8 storage (quantized values), pVect2 = FP32 query
+static inline void InnerProductStepSQ8(const uint8_t *pVect1, const float *pVect2, size_t &offset,


Suggested change

static inline void InnerProductStepSQ8(const uint8_t *pVect1, const float *pVect2, size_t &offset,

static inline void InnerProductStepSQ8_FP32(const uint8_t *pVect1, const float *pVect2, size_t &offset,

meiravgri · 2026-01-13T14:35:53Z

src/VecSim/spaces/functions/AVX512F.h

+dist_func_t<float> Choose_SQ8_FP32_IP_implementation_AVX512F(size_t dim);
+dist_func_t<float> Choose_SQ8_FP32_Cosine_implementation_AVX512F(size_t dim);


does these exist?

nope, not sure why it was added

meiravgri · 2026-01-13T14:38:30Z

tests/unit/test_spaces.cpp

 /* ======================== Tests SQ8_SQ8 ========================= */

-TEST_F(SpacesTest, SQ8_SQ8_ip_no_optimization_func_test) {
+TEST_F(SpacesTest, SQ8_SQ8_FP32_ip_no_optimization_func_test) {


he has a point

… implementations

Copilot

Pull request overview

Copilot reviewed 42 out of 42 changed files in this pull request and generated no new comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

dor-forer requested a review from Copilot January 12, 2026 13:22

Copilot started reviewing on behalf of dor-forer January 12, 2026 13:22 View session

Copilot AI reviewed Jan 12, 2026

View reviewed changes

dor-forer added 2 commits January 12, 2026 15:28

Rename SQ8 header files to SQ8_FP32

2bdb6f1

- Rename IP headers: IP_*_SQ8.h -> IP_*_SQ8_FP32.h - Rename L2 headers: L2_*_SQ8.h -> L2_*_SQ8_FP32.h - Rename benchmark file: bm_spaces_sq8.cpp -> bm_spaces_sq8_fp32.cpp

dor-forer force-pushed the dorer-rename-to-sq8-fp32 branch from d048ec4 to 2bdb6f1 Compare January 12, 2026 13:30

dor-forer requested a review from Copilot January 12, 2026 13:30

Copilot started reviewing on behalf of dor-forer January 12, 2026 13:30 View session

Copilot AI reviewed Jan 12, 2026

View reviewed changes

dor-forer added the bm-spaces-sq8-full label Jan 12, 2026

dor-forer requested a review from meiravgri January 12, 2026 13:42

meiravgri reviewed Jan 13, 2026

View reviewed changes

Rename SQ8 functions and variables to SQ8_FP32 for consistency across…

18c1a81

… implementations

dor-forer requested review from Copilot and meiravgri January 14, 2026 07:00

Copilot started reviewing on behalf of dor-forer January 14, 2026 07:00 View session

Copilot AI reviewed Jan 14, 2026

View reviewed changes

dor-forer enabled auto-merge January 14, 2026 08:42

meiravgri approved these changes Jan 14, 2026

View reviewed changes

dor-forer added this pull request to the merge queue Jan 14, 2026

Merged via the queue into main with commit 7e1a78c Jan 14, 2026
30 checks passed

dor-forer deleted the dorer-rename-to-sq8-fp32 branch January 14, 2026 09:25

	static inline void InnerProductStepSQ8(const uint8_t &pVect1, const float &pVect2,
	static inline void InnerProductStepSQ8_FP32(const uint8_t &pVect1, const float &pVect2,

	static inline void SQ8_InnerProductStep(const uint8_t &pVec1, const float &pVec2, __m512 &sum) {
	static inline void SQ8_FP32_InnerProductStep(const uint8_t &pVec1, const float &pVec2, __m512 &sum) {

		dist_func_t<float> Choose_SQ8_FP32_IP_implementation_AVX512F(size_t dim);
		dist_func_t<float> Choose_SQ8_FP32_Cosine_implementation_AVX512F(size_t dim);

Conversation

dor-forer commented Jan 12, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI Jan 12, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Copilot AI Jan 12, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

codecov bot commented Jan 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

meiravgri left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dor-forer Jan 14, 2026 • 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!

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

codecov bot commented Jan 12, 2026 •

edited

Loading

dor-forer Jan 14, 2026 •

edited

Loading