Jaeyoon Jeong e7c44eb000 Add AVX-512 fp16 conversion kernels (#5694) ...

Summary:
Pull Request resolved: https://github.com/facebookresearch/faiss/pull/5694

Add exact AVX-512 widening and fp32 accumulation specializations for packed fp16 data. Reuse the existing AVX-512 SIMD translation unit, expand runtime dispatch to the AVX-512 level, and cover conversion tails, large blocks, weighted and unweighted centroid updates, and architecture fallbacks.

Reviewed By: mnorris11

Differential Revision: D122781509

fbshipit-source-id: 67cbc4546149f069706e80dd25cae25d3588b0cc

2026-10-02 18:24:52 -07:00

..

index_io_backward_compatibility

Fix cuVS GPU CI timeout: pin CUDA toolkit to 13.2 (#5401)

2026-07-10 17:33:47 -07:00

bench_hnsw_deterministic.py

faiss HNSW: delete the flag and the lock-based graph build (#5611)

2026-09-14 13:13:11 -07:00

bench_transposed_encode.py

De-flake bench_transposed_encode (CI timeout) (#5360)

2026-06-29 15:07:10 -07:00

CMakeLists.txt

RaBitQ: add batch-four estimates and 9-bit SIMD scoring (#5647)

2026-10-02 07:58:02 -07:00

common_faiss_tests.py

Add AVX512 VPOPCNT dynamic dispatch level (#5531)

2026-09-10 09:32:36 -07:00

external_module_test.py

Re-add example of how to build, link, and test an external SWIG module"" (#3981)

2024-10-23 07:34:38 -07:00

test_approx_topk.cpp

Convert approx_topk and residual_quantizer_encode_steps to SIMDLevel dynamic dispatch (#4923)

2026-03-14 15:12:14 -07:00

test_autotune.py

Fix OperatingPoints::add prune loop leaving dominated points at equal time (#5415)

2026-07-13 11:09:13 -07:00

test_binary_autotune.py

support SearchParameters for IndexBinary (#4761)

2026-02-04 00:03:15 -08:00

test_binary_factory.py

Add IndexBinaryIDMap2 support to index binary factory (#4603)

2025-10-14 16:47:57 -07:00

test_binary_flat.cpp

Enable linting: lint config changes plus arc lint command (#3966)

2024-10-22 09:46:48 -07:00

test_binary_hash.cpp

Fix IndexBinaryMultiHash::reset() not clearing hash maps and avoid unnecessary copies (#5100)

2026-04-14 23:35:31 -07:00

test_binary_hashindex.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_binary_io.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_binary_ivf_scanner.py

Measure 20-byte Hamming codes eight at a time (#5587)

2026-09-10 09:32:36 -07:00

test_binary_search_params.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_build_blocks.py

black -l 80: tests/ formatting (2/3) (#5427)

2026-07-15 17:47:15 -07:00

test_callback_py.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_callback.cpp

Reduce memory usage in timeout callback tests (#4745)

2026-01-05 15:00:20 -08:00

test_clone.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_clustering_initialization.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_clustering.py

Finish E501 line-length lint sweep across faiss (183 violations, 52 files) (#5444)

2026-07-17 10:20:42 -07:00

test_common_ivf_empty_index.cpp

fix: resolve compiler warnings in test files (#4944)

2026-03-20 10:21:55 -07:00

test_contrib_with_scipy.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_contrib.py

Fix reverse_index_factory/get_code_size round-trip for HNSW, IMI and IVFPQR (#5564)

2026-09-04 11:14:06 -07:00

test_cppcontrib_sa_decode.cpp

Fix CQS signal readability-braces-around-statements in fbcode/faiss/tests (#4457)

2025-07-25 02:37:10 -07:00

test_cppcontrib_uintreader.cpp

fix: resolve compiler warnings in test files (#4944)

2026-03-20 10:21:55 -07:00

test_custom_result_handler.cpp

Add staged RaBitQ storage to HNSW (#5526)

2026-09-08 08:53:40 -07:00

test_dealloc_invlists.cpp

fix: resolve compiler warnings in test files (#4944)

2026-03-20 10:21:55 -07:00

test_disable_pq_sdc_tables.cpp

fix: resolve compiler warnings in test files (#4944)

2026-03-20 10:21:55 -07:00

test_distances_dispatch.cpp

Add AVX-512 fp16 conversion kernels (#5694)

2026-10-02 18:24:52 -07:00

test_distances_simd.cpp

ARM SVE optimization for L2 distance kernels adapted to new simd_impl structure (#5098)

2026-08-25 10:20:33 -07:00

test_documentation.py

Enable linting: lint config changes plus arc lint command (#3966)

2024-10-22 09:46:48 -07:00

test_eden.py

Clamp EDEN unbiased L2 distances to be non-negative (#5569)

2026-09-04 04:34:19 -07:00

test_extra_distances.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_factory_tools.cpp

Fix reverse_index_factory for multi-bit RaBitQ indexes (#5419)

2026-07-14 12:27:43 -07:00

test_factory.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_fast_scan_distance_to_code.cpp

Implement distance_to_code for IVFRaBitQFastScanScanner (#4822)

2026-02-20 19:39:13 -08:00

test_fast_scan_ivf.py

Fix division-by-zero in quantize_lut when all LUT entries are identical (#5424)

2026-07-15 08:50:29 -07:00

test_fast_scan.py

Bound the fast scan LUT scale by the accumulated sum (#5658)

2026-09-25 13:47:19 -07:00

test_fastscan_filter.cpp

introduce SingleQueryResultCollectHandler (#4926)

2026-03-18 14:05:55 -07:00

test_fastscan_filter.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_fastscan_perf.cpp

Enable dynamic dispatch build on Windows (#5127)

2026-04-23 13:31:11 -07:00

test_flat_panorama.py

De-flake test_flat_panorama across SIMD levels (#5361)

2026-06-30 10:17:47 -07:00

test_graph_based.py

Make search stats collection configurable (#5670)

2026-09-26 21:59:14 -07:00

test_hamming_utils.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_hamming.cpp

Accept ragged code sizes in faiss::hammings(), and fix the vectorization axis for wide codes (#5520)

2026-08-19 10:15:02 -07:00

test_heap.cpp

faiss: initialize ids_tab to -1 in Top1BlockResultHandler::begin_multiple (#5249)

2026-05-28 13:28:59 -07:00

test_hnsw_panorama.py

Finish E501 line-length lint sweep across faiss (183 violations, 52 files) (#5444)

2026-07-17 10:20:42 -07:00

test_hnsw_rabitq.py

Make search stats collection configurable (#5670)

2026-09-26 21:59:14 -07:00

test_hnsw.cpp

Avoid double-advancing HNSW visited tables (#5596)

2026-10-01 10:36:44 -07:00

test_index_accuracy.py

Stop aarch64 falling back to SIMDLevel::NONE in faiss dispatch (#5572)

2026-09-10 09:32:36 -07:00

test_index_binary_from_float.py

Fix distance conversion in IndexBinaryFromFloat for non-L2 metrics (#5549)

2026-09-01 10:47:31 -07:00

test_index_binary.py

Fix IndexBinaryIVF reconstruct_n and search_and_reconstruct byte stride (#5463)

2026-08-04 11:03:23 -07:00

test_index_composite.py

Use unittest assertions and f-strings in test_index_composite (#5571)

2026-09-08 14:08:01 -07:00

test_index.py

Fix empty-index BLAS search results (#5528)

2026-08-25 07:56:20 -07:00

test_io_corrupted.cpp

Fix NSG off-by-one neighbor ID check (#4804)

2026-02-12 21:09:31 -08:00

test_io.py

Add staged RaBitQ storage to HNSW (#5526)

2026-09-08 08:53:40 -07:00

test_ivf_early_termination.cpp

facebook-unused-include-check in test_ivf_early_termination.cpp (#5376)

2026-07-06 08:54:27 -07:00

test_ivf_flat_panorama.py

Update default Panorama batch size to 1024 (#5441)

2026-08-03 10:53:18 -07:00

test_ivf_index.cpp

Avoid widening fp16 data for k-means++ initialization (#5693)

2026-10-02 18:24:52 -07:00

test_ivf_sharding.cpp

Fix two memory leaks in shard_ivf_index_centroids (#5495)

2026-07-31 20:58:26 -07:00

test_ivf_sq_fastscan.py

Add fast scan variants of SQ indexes - IndexSQFastScan and IndexIVFSQFastScan (#5341)

2026-09-18 20:24:54 -07:00

test_ivflib.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_ivfpq_codec.cpp

facebook-unused-include-check in fbcode/faiss (#4029)

2024-11-15 13:53:46 -08:00

test_ivfpq_indexing.cpp

Add explicit IVFPQ scanner distance modes (#5632)

2026-09-15 17:32:23 -07:00

test_local_search_quantizer.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_lowlevel_ivf.cpp

fix: resolve compiler warnings in test files (#4944)

2026-03-20 10:21:55 -07:00

test_merge_index.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_merge.cpp

fix: resolve compiler warnings in test files (#4944)

2026-03-20 10:21:55 -07:00

test_meta_index.py

Merge shard results by metric type rather than assuming similarity (#5509)

2026-08-26 12:48:20 -07:00

test_metal_ivfpq.py

Add GPU-accelerated IVFPQ search to the Metal backend (#5449)

2026-10-01 21:07:27 -07:00

test_mmap.cpp

Enable dynamic dispatch build on Windows (#5127)

2026-04-23 13:31:11 -07:00

test_nndescent.cpp

Fix correctness bugs in NNDescent Nhood copy/move operations and gen_random bounds (#5072)

2026-04-09 20:12:31 -07:00

test_NSG_compressed_graph.cpp

avoid runtime checks in VisitedTable (#5234)

2026-05-30 02:02:42 -07:00

test_omp_exception_safety.cpp

Fix OMP exception safety in IndexHNSW search (#5133)

2026-04-24 12:53:56 -07:00

test_omp_threads_py.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_omp_threads.cpp

Enable linting: lint config changes plus arc lint command (#3966)

2024-10-22 09:46:48 -07:00

test_ondisk_ivf.cpp

Fix typos in tests and contrib directories (#4672)

2025-11-12 11:39:51 -08:00

test_pairs_decoding.cpp

fix: resolve compiler warnings in test files (#4944)

2026-03-20 10:21:55 -07:00

test_params_override.cpp

IDSelectorWithContext: extend context hook to IVFPQ/PQR, multibit RaBitQ, and Panorama scanners (#5491)

2026-08-05 10:36:30 -07:00

test_partition.py

Fix genuine flake8 lint in faiss Python (F401/F403/F841/B011) (#5351)

2026-06-29 18:38:01 -07:00

test_partitioning.cpp

Work around GCC 12 miscompilation of AVX2 histogram (#5124)

2026-05-05 00:46:37 -07:00

test_pq_code_distance.cpp

Inline PQ code distance kernels into scanner TUs (#5159)

2026-05-04 07:35:56 -07:00

test_pq_encoding.cpp

Add FastScanCodeScanner dispatch boundary with per-SIMD TUs (#4897)

2026-03-12 02:15:42 -07:00

test_pqfs_unaligned.cpp

Enable dynamic dispatch build on Windows (#5127)

2026-04-23 13:31:11 -07:00

test_product_quantizer.py

Add RVV SIMD specialization for product quantizer distance-table computation (#5543)

2026-09-24 18:31:50 -07:00

test_rabitq_batch_estimates.cpp

RaBitQ: add batch-four estimates and 9-bit SIMD scoring (#5647)

2026-10-02 07:58:02 -07:00

test_rabitq_fastscan.cpp

Fix RaBitQ fast scan aux offsets for large bbs (#5421)

2026-07-31 07:49:57 -07:00

test_rabitq_fastscan.py

Fix RaBitQ fast scan aux offsets for large bbs (#5421)

2026-07-31 07:49:57 -07:00

test_rabitq_simd_dd.cpp

rabitq neon kernels (#5532)

2026-10-01 10:36:44 -07:00

test_rabitq_simd_neon.cpp

rabitq neon kernels (#5532)

2026-10-01 10:36:44 -07:00

test_rabitq_simd_util.h

rabitq neon kernels (#5532)

2026-10-01 10:36:44 -07:00

test_rabitq_simd.cpp

rabitq neon kernels (#5532)

2026-10-01 10:36:44 -07:00

test_rabitq_utils.py

Port test_rabitq_utils from C++ to Python (#5442)

2026-07-17 11:14:42 -07:00

test_rabitq.py

RaBitQ: Fix zero residual sign convention (#5553)

2026-09-03 09:56:13 -07:00

test_RCQ_cropping.cpp

fix: resolve compiler warnings in test files (#4944)

2026-03-20 10:21:55 -07:00

test_read_index_deserialize.cpp

Make Panorama bounds checks overflow-safe (#5616)

2026-09-14 21:26:48 -07:00

test_referenced_objects.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_refine_panorama.py

Fix panorama tests: pass D_panorama (not D_regular) to check_ref_knn_with_draws (#5430)

2026-07-16 12:09:45 -07:00

test_refine.py

Finish E501 line-length lint sweep across faiss (183 violations, 52 files) (#5444)

2026-07-17 10:20:42 -07:00

test_residual_quantizer.py

Fix: Windows SIMD detection (#5497)

2026-09-04 15:27:56 -07:00

test_rowwise_minmax.py

Enable linting: lint config changes plus arc lint command (#3966)

2024-10-22 09:46:48 -07:00

test_scalar_quantizer_correctness.py

Run the eight SIMD kernel tests that no build executed (#5620)

2026-09-10 20:05:56 -07:00

test_scalar_quantizer.cpp

Add RVV encoding for QT_8bit_direct (#5640)

2026-09-24 12:35:28 -07:00

test_search_params.py

Fix IDSelector handling in unbounded HNSW search (#5508)

2026-09-16 17:55:51 -07:00

test_simd_dispatch.py

Measure 20-byte Hamming codes eight at a time (#5587)

2026-09-10 09:32:36 -07:00

test_simd_levels_x86_avx2.cpp

Introduce Dynamic Dispatch infrastructure with SIMDConfig (#4780)

2026-02-09 01:19:35 -08:00

test_simd_levels_x86_avx512.cpp

Introduce Dynamic Dispatch infrastructure with SIMDConfig (#4780)

2026-02-09 01:19:35 -08:00

test_simd_levels.cpp

Add AVX512 VPOPCNT dynamic dispatch level (#5531)

2026-09-10 09:32:36 -07:00

test_simd_perf.cpp

Remove flaky test (#5169)

2026-05-01 09:55:55 -07:00

test_simdlib.cpp

Remove bare simdlib type aliases from simdlib_dispatch.h (#5069)

2026-04-09 04:19:54 -07:00

test_single_query_collect_handler.cpp

introduce SingleQueryResultCollectHandler (#4926)

2026-03-18 14:05:55 -07:00

test_sliding_ivf.cpp

fix: resolve compiler warnings in test files (#4944)

2026-03-20 10:21:55 -07:00

test_sorting.cpp

Fix race condition, memory management, debug output, and hashtable lookup in sorting.cpp (#5078)

2026-04-10 17:59:30 -07:00

test_sq_fastscan.py

Add reranked types to IndexSQFastScan and fix flat RaBitQ filtering (#5661)

2026-09-25 13:47:19 -07:00

test_sq_rvv_tail_and_semantics.cpp

Add RVV SIMD specialization for scalar quantizer distance compute (#5539)

2026-09-17 13:34:43 -07:00

test_standalone_codec.py

Finish E501 line-length lint sweep across faiss (183 violations, 52 files) (#5444)

2026-07-17 10:20:42 -07:00

test_super_kmeans_foundations.cpp

Extract super_kmeans_assign_iteration as a shared public SuperKMeans kernel (#5443)

2026-07-17 21:15:37 -07:00

test_super_kmeans.py

Add SVE low-dimensional nearest fast path and spherical SuperKMeans support (#5530)

2026-09-02 07:56:05 -07:00

test_svs_metric.cpp

Fix deadly signal in to_svs_metric() for unsupported metrics (#5032)

2026-04-03 15:08:59 -07:00

test_svs_py.py

SVS: Add ParameterSpace.set_index_parameter support for Vamana (#5644)

2026-09-18 10:15:10 -07:00

test_svs.cpp

SVS: allow option to disable vector storing (#5525)

2026-09-04 10:02:36 -07:00

test_swig_wrapper.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_threaded_index.cpp

Remove confusing double-negative throw-macro conditions (#5471)

2026-07-27 09:25:51 -07:00

test_transfer_invlists.cpp

fix: resolve compiler warnings in test files (#4944)

2026-03-20 10:21:55 -07:00

test_util.h

fix: resolve compiler warnings in test files (#4944)

2026-03-20 10:21:55 -07:00

test_utils.cpp

Enable linting: lint config changes plus arc lint command (#3966)

2024-10-22 09:46:48 -07:00

test_wheel_smoke_gpu.py

Pin faiss-gpu-cuvs RAPIDS deps to the 26.06 build series (#5643)

2026-09-15 14:42:20 -07:00

test_wheel_smoke.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

test_zerocopy.cpp

Support ZeroCopyIOReader for IndexRaBitQ deserialization (#5629)

2026-09-14 13:03:13 -07:00

torch_test_contrib.py

Format faiss Python files with black (line length 80) (#5349)

2026-06-29 17:29:05 -07:00

torch_test_neural_net.py

Finish E501 line-length lint sweep across faiss (183 violations, 52 files) (#5444)

2026-07-17 10:20:42 -07:00