faiss 0.6.0 → 0.6.1

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: faiss
 version: !ruby/object:Gem::Version
-  version: 0.6.0
+  version: 0.6.1
 platform: ruby
 authors:
 - Andrew Kane
@@ -48,14 +48,14 @@ files:
 - README.md
 - ext/faiss/ext.cpp
 - ext/faiss/extconf.rb
+- ext/faiss/index.cpp
 - ext/faiss/index_binary.cpp
-- ext/faiss/index_rb.cpp
 - ext/faiss/kmeans.cpp
 - ext/faiss/numo.hpp
 - ext/faiss/pca_matrix.cpp
 - ext/faiss/product_quantizer.cpp
-- ext/faiss/utils_rb.cpp
-- ext/faiss/utils_rb.h
+- ext/faiss/utils.cpp
+- ext/faiss/utils.h
 - lib/faiss.rb
 - lib/faiss/version.rb
 - vendor/faiss/LICENSE
@@ -158,8 +158,12 @@ files:
 - vendor/faiss/faiss/MetaIndexes.cpp
 - vendor/faiss/faiss/MetaIndexes.h
 - vendor/faiss/faiss/MetricType.h
+- vendor/faiss/faiss/SuperKMeans.cpp
+- vendor/faiss/faiss/SuperKMeans.h
 - vendor/faiss/faiss/VectorTransform.cpp
 - vendor/faiss/faiss/VectorTransform.h
+- vendor/faiss/faiss/build.cpp
+- vendor/faiss/faiss/build.h
 - vendor/faiss/faiss/clone_index.cpp
 - vendor/faiss/faiss/clone_index.h
 - vendor/faiss/faiss/cppcontrib/SaDecodeKernels.h
@@ -227,10 +231,22 @@ files:
 - vendor/faiss/faiss/gpu/utils/StaticUtils.h
 - vendor/faiss/faiss/gpu/utils/Timer.cpp
 - vendor/faiss/faiss/gpu/utils/Timer.h
+- vendor/faiss/faiss/gpu_metal/GpuIndexFlat.h
+- vendor/faiss/faiss/gpu_metal/MetalCloner.h
+- vendor/faiss/faiss/gpu_metal/MetalFlatKernels.h
+- vendor/faiss/faiss/gpu_metal/MetalIndex.h
+- vendor/faiss/faiss/gpu_metal/MetalIndexFlat.h
+- vendor/faiss/faiss/gpu_metal/MetalKernels.h
+- vendor/faiss/faiss/gpu_metal/MetalResources.h
+- vendor/faiss/faiss/gpu_metal/StandardMetalResources.h
+- vendor/faiss/faiss/impl/AdSampling.cpp
+- vendor/faiss/faiss/impl/AdSampling.h
 - vendor/faiss/faiss/impl/AdditiveQuantizer.cpp
 - vendor/faiss/faiss/impl/AdditiveQuantizer.h
 - vendor/faiss/faiss/impl/AuxIndexStructures.cpp
 - vendor/faiss/faiss/impl/AuxIndexStructures.h
+- vendor/faiss/faiss/impl/ClusteringHelpers.cpp
+- vendor/faiss/faiss/impl/ClusteringHelpers.h
 - vendor/faiss/faiss/impl/ClusteringInitialization.cpp
 - vendor/faiss/faiss/impl/ClusteringInitialization.h
 - vendor/faiss/faiss/impl/CodePacker.cpp
@@ -241,14 +257,13 @@ files:
 - vendor/faiss/faiss/impl/FaissAssert.h
 - vendor/faiss/faiss/impl/FaissException.cpp
 - vendor/faiss/faiss/impl/FaissException.h
-- vendor/faiss/faiss/impl/FastScanDistancePostProcessing.h
 - vendor/faiss/faiss/impl/HNSW.cpp
 - vendor/faiss/faiss/impl/HNSW.h
 - vendor/faiss/faiss/impl/IDSelector.cpp
 - vendor/faiss/faiss/impl/IDSelector.h
+- vendor/faiss/faiss/impl/InvertedListScannerStats.h
 - vendor/faiss/faiss/impl/LocalSearchQuantizer.cpp
 - vendor/faiss/faiss/impl/LocalSearchQuantizer.h
-- vendor/faiss/faiss/impl/LookupTableScaler.h
 - vendor/faiss/faiss/impl/NNDescent.cpp
 - vendor/faiss/faiss/impl/NNDescent.h
 - vendor/faiss/faiss/impl/NSG.cpp
@@ -257,6 +272,8 @@ files:
 - vendor/faiss/faiss/impl/Panorama.h
 - vendor/faiss/faiss/impl/PanoramaStats.cpp
 - vendor/faiss/faiss/impl/PanoramaStats.h
+- vendor/faiss/faiss/impl/PdxLayout.cpp
+- vendor/faiss/faiss/impl/PdxLayout.h
 - vendor/faiss/faiss/impl/PolysemousTraining.cpp
 - vendor/faiss/faiss/impl/PolysemousTraining.h
 - vendor/faiss/faiss/impl/ProductAdditiveQuantizer.cpp
@@ -265,8 +282,6 @@ files:
 - vendor/faiss/faiss/impl/ProductQuantizer.cpp
 - vendor/faiss/faiss/impl/ProductQuantizer.h
 - vendor/faiss/faiss/impl/Quantizer.h
-- vendor/faiss/faiss/impl/RaBitQStats.cpp
-- vendor/faiss/faiss/impl/RaBitQStats.h
 - vendor/faiss/faiss/impl/RaBitQUtils.cpp
 - vendor/faiss/faiss/impl/RaBitQUtils.h
 - vendor/faiss/faiss/impl/RaBitQuantizer.cpp
@@ -282,7 +297,47 @@ files:
 - vendor/faiss/faiss/impl/ThreadedIndex.h
 - vendor/faiss/faiss/impl/VisitedTable.cpp
 - vendor/faiss/faiss/impl/VisitedTable.h
+- vendor/faiss/faiss/impl/approx_topk/approx_topk.h
+- vendor/faiss/faiss/impl/approx_topk/avx2.cpp
+- vendor/faiss/faiss/impl/approx_topk/generic.h
+- vendor/faiss/faiss/impl/approx_topk/neon.cpp
+- vendor/faiss/faiss/impl/approx_topk/rq_beam_search_tab-inl.h
+- vendor/faiss/faiss/impl/approx_topk/rq_beam_search_tab.h
+- vendor/faiss/faiss/impl/approx_topk/simdlib256-inl.h
+- vendor/faiss/faiss/impl/binary_hamming/IndexBinaryHNSW_impl.h
+- vendor/faiss/faiss/impl/binary_hamming/IndexBinaryHash_impl.h
+- vendor/faiss/faiss/impl/binary_hamming/IndexBinaryIVF_impl.h
+- vendor/faiss/faiss/impl/binary_hamming/IndexIVFSpectralHash_impl.h
+- vendor/faiss/faiss/impl/binary_hamming/IndexPQ_impl.h
+- vendor/faiss/faiss/impl/binary_hamming/avx2.cpp
+- vendor/faiss/faiss/impl/binary_hamming/avx512.cpp
+- vendor/faiss/faiss/impl/binary_hamming/dispatch.h
+- vendor/faiss/faiss/impl/binary_hamming/neon.cpp
+- vendor/faiss/faiss/impl/binary_hamming/rvv.cpp
 - vendor/faiss/faiss/impl/expanded_scanners.h
+- vendor/faiss/faiss/impl/fast_scan/FastScanDistancePostProcessing.h
+- vendor/faiss/faiss/impl/fast_scan/LookupTableScaler.h
+- vendor/faiss/faiss/impl/fast_scan/accumulate_loops.h
+- vendor/faiss/faiss/impl/fast_scan/accumulate_loops_512.h
+- vendor/faiss/faiss/impl/fast_scan/decompose_qbs.h
+- vendor/faiss/faiss/impl/fast_scan/dispatching.h
+- vendor/faiss/faiss/impl/fast_scan/fast_scan.cpp
+- vendor/faiss/faiss/impl/fast_scan/fast_scan.h
+- vendor/faiss/faiss/impl/fast_scan/impl-avx2.cpp
+- vendor/faiss/faiss/impl/fast_scan/impl-avx512.cpp
+- vendor/faiss/faiss/impl/fast_scan/impl-neon.cpp
+- vendor/faiss/faiss/impl/fast_scan/impl-riscv.cpp
+- vendor/faiss/faiss/impl/fast_scan/kernels_simd256.h
+- vendor/faiss/faiss/impl/fast_scan/kernels_simd512.h
+- vendor/faiss/faiss/impl/fast_scan/rabitq_dispatching.h
+- vendor/faiss/faiss/impl/fast_scan/rabitq_result_handler.h
+- vendor/faiss/faiss/impl/fast_scan/simd_result_handlers.h
+- vendor/faiss/faiss/impl/hnsw/LockVector.cpp
+- vendor/faiss/faiss/impl/hnsw/LockVector.h
+- vendor/faiss/faiss/impl/hnsw/MinimaxHeap.cpp
+- vendor/faiss/faiss/impl/hnsw/MinimaxHeap.h
+- vendor/faiss/faiss/impl/hnsw/avx2.cpp
+- vendor/faiss/faiss/impl/hnsw/avx512.cpp
 - vendor/faiss/faiss/impl/index_read.cpp
 - vendor/faiss/faiss/impl/index_read_utils.h
 - vendor/faiss/faiss/impl/index_write.cpp
@@ -297,15 +352,20 @@ files:
 - vendor/faiss/faiss/impl/mapped_io.h
 - vendor/faiss/faiss/impl/maybe_owned_vector.h
 - vendor/faiss/faiss/impl/platform_macros.h
-- vendor/faiss/faiss/impl/pq4_fast_scan.cpp
-- vendor/faiss/faiss/impl/pq4_fast_scan.h
-- vendor/faiss/faiss/impl/pq4_fast_scan_search_1.cpp
-- vendor/faiss/faiss/impl/pq4_fast_scan_search_qbs.cpp
-- vendor/faiss/faiss/impl/pq_code_distance/pq_code_distance-avx2.cpp
-- vendor/faiss/faiss/impl/pq_code_distance/pq_code_distance-avx512.cpp
+- vendor/faiss/faiss/impl/pq_code_distance/IVFPQScanner_impl.h
+- vendor/faiss/faiss/impl/pq_code_distance/IVFPQ_QueryTables.cpp
+- vendor/faiss/faiss/impl/pq_code_distance/IVFPQ_QueryTables.h
+- vendor/faiss/faiss/impl/pq_code_distance/PQDistanceComputer_impl.h
+- vendor/faiss/faiss/impl/pq_code_distance/avx2.cpp
+- vendor/faiss/faiss/impl/pq_code_distance/avx512.cpp
+- vendor/faiss/faiss/impl/pq_code_distance/neon.cpp
+- vendor/faiss/faiss/impl/pq_code_distance/pq_code_distance-avx2.h
+- vendor/faiss/faiss/impl/pq_code_distance/pq_code_distance-avx512.h
 - vendor/faiss/faiss/impl/pq_code_distance/pq_code_distance-generic.cpp
+- vendor/faiss/faiss/impl/pq_code_distance/pq_code_distance-generic.h
 - vendor/faiss/faiss/impl/pq_code_distance/pq_code_distance-inl.h
 - vendor/faiss/faiss/impl/pq_code_distance/pq_code_distance-sve.cpp
+- vendor/faiss/faiss/impl/pq_code_distance/rvv.cpp
 - vendor/faiss/faiss/impl/residual_quantizer_encode_steps.cpp
 - vendor/faiss/faiss/impl/residual_quantizer_encode_steps.h
 - vendor/faiss/faiss/impl/scalar_quantizer/codecs.h
@@ -317,10 +377,17 @@ files:
 - vendor/faiss/faiss/impl/scalar_quantizer/sq-avx512.cpp
 - vendor/faiss/faiss/impl/scalar_quantizer/sq-dispatch.h
 - vendor/faiss/faiss/impl/scalar_quantizer/sq-neon.cpp
+- vendor/faiss/faiss/impl/scalar_quantizer/sq-rvv.cpp
 - vendor/faiss/faiss/impl/scalar_quantizer/training.cpp
 - vendor/faiss/faiss/impl/scalar_quantizer/training.h
 - vendor/faiss/faiss/impl/simd_dispatch.h
-- vendor/faiss/faiss/impl/simd_result_handlers.h
+- vendor/faiss/faiss/impl/simdlib/simdlib.h
+- vendor/faiss/faiss/impl/simdlib/simdlib_avx2.h
+- vendor/faiss/faiss/impl/simdlib/simdlib_avx512.h
+- vendor/faiss/faiss/impl/simdlib/simdlib_dispatch.h
+- vendor/faiss/faiss/impl/simdlib/simdlib_emulated.h
+- vendor/faiss/faiss/impl/simdlib/simdlib_neon.h
+- vendor/faiss/faiss/impl/simdlib/simdlib_ppc64.h
 - vendor/faiss/faiss/impl/svs_io.cpp
 - vendor/faiss/faiss/impl/svs_io.h
 - vendor/faiss/faiss/impl/zerocopy_io.cpp
@@ -343,6 +410,12 @@ files:
 - vendor/faiss/faiss/svs/IndexSVSFaissUtils.h
 - vendor/faiss/faiss/svs/IndexSVSFlat.cpp
 - vendor/faiss/faiss/svs/IndexSVSFlat.h
+- vendor/faiss/faiss/svs/IndexSVSIVF.cpp
+- vendor/faiss/faiss/svs/IndexSVSIVF.h
+- vendor/faiss/faiss/svs/IndexSVSIVFLVQ.cpp
+- vendor/faiss/faiss/svs/IndexSVSIVFLVQ.h
+- vendor/faiss/faiss/svs/IndexSVSIVFLeanVec.cpp
+- vendor/faiss/faiss/svs/IndexSVSIVFLeanVec.h
 - vendor/faiss/faiss/svs/IndexSVSVamana.cpp
 - vendor/faiss/faiss/svs/IndexSVSVamana.h
 - vendor/faiss/faiss/svs/IndexSVSVamanaLVQ.cpp
@@ -356,23 +429,18 @@ files:
 - vendor/faiss/faiss/utils/NeuralNet.h
 - vendor/faiss/faiss/utils/WorkerThread.cpp
 - vendor/faiss/faiss/utils/WorkerThread.h
-- vendor/faiss/faiss/utils/approx_topk/approx_topk.h
-- vendor/faiss/faiss/utils/approx_topk/avx2-inl.h
-- vendor/faiss/faiss/utils/approx_topk/generic.h
-- vendor/faiss/faiss/utils/approx_topk/mode.h
 - vendor/faiss/faiss/utils/approx_topk_hamming/approx_topk_hamming.h
 - vendor/faiss/faiss/utils/bf16.h
 - vendor/faiss/faiss/utils/distances.cpp
 - vendor/faiss/faiss/utils/distances.h
 - vendor/faiss/faiss/utils/distances_dispatch.h
 - vendor/faiss/faiss/utils/distances_fused/avx512.cpp
-- vendor/faiss/faiss/utils/distances_fused/avx512.h
 - vendor/faiss/faiss/utils/distances_fused/distances_fused.cpp
 - vendor/faiss/faiss/utils/distances_fused/distances_fused.h
 - vendor/faiss/faiss/utils/distances_fused/simdlib_based.cpp
-- vendor/faiss/faiss/utils/distances_fused/simdlib_based.h
+- vendor/faiss/faiss/utils/distances_fused/simdlib_based_neon.cpp
+- vendor/faiss/faiss/utils/distances_fused/simdlib_kernel-inl.h
 - vendor/faiss/faiss/utils/distances_simd.cpp
-- vendor/faiss/faiss/utils/extra_distances-inl.h
 - vendor/faiss/faiss/utils/extra_distances.cpp
 - vendor/faiss/faiss/utils/extra_distances.h
 - vendor/faiss/faiss/utils/fp16-arm.h
@@ -382,15 +450,22 @@ files:
 - vendor/faiss/faiss/utils/hamming-inl.h
 - vendor/faiss/faiss/utils/hamming.cpp
 - vendor/faiss/faiss/utils/hamming.h
-- vendor/faiss/faiss/utils/hamming_distance/avx2-inl.h
-- vendor/faiss/faiss/utils/hamming_distance/avx512-inl.h
 - vendor/faiss/faiss/utils/hamming_distance/common.h
-- vendor/faiss/faiss/utils/hamming_distance/generic-inl.h
-- vendor/faiss/faiss/utils/hamming_distance/hamdis-inl.h
-- vendor/faiss/faiss/utils/hamming_distance/neon-inl.h
+- vendor/faiss/faiss/utils/hamming_distance/hamming_avx2.cpp
+- vendor/faiss/faiss/utils/hamming_distance/hamming_avx512.cpp
+- vendor/faiss/faiss/utils/hamming_distance/hamming_computer-avx2.h
+- vendor/faiss/faiss/utils/hamming_distance/hamming_computer-avx512.h
+- vendor/faiss/faiss/utils/hamming_distance/hamming_computer-generic.h
+- vendor/faiss/faiss/utils/hamming_distance/hamming_computer-neon.h
+- vendor/faiss/faiss/utils/hamming_distance/hamming_computer-rvv.h
+- vendor/faiss/faiss/utils/hamming_distance/hamming_computer.h
+- vendor/faiss/faiss/utils/hamming_distance/hamming_impl.h
+- vendor/faiss/faiss/utils/hamming_distance/hamming_neon.cpp
+- vendor/faiss/faiss/utils/hamming_distance/hamming_rvv.cpp
 - vendor/faiss/faiss/utils/ordered_key_value.h
 - vendor/faiss/faiss/utils/partitioning.cpp
 - vendor/faiss/faiss/utils/partitioning.h
+- vendor/faiss/faiss/utils/popcount.h
 - vendor/faiss/faiss/utils/pq_code_distance.h
 - vendor/faiss/faiss/utils/prefetch.h
 - vendor/faiss/faiss/utils/quantize_lut.cpp
@@ -398,20 +473,30 @@ files:
 - vendor/faiss/faiss/utils/rabitq_simd.h
 - vendor/faiss/faiss/utils/random.cpp
 - vendor/faiss/faiss/utils/random.h
+- vendor/faiss/faiss/utils/simd_impl/IVFFlatScanner-inl.h
 - vendor/faiss/faiss/utils/simd_impl/distances_aarch64.cpp
 - vendor/faiss/faiss/utils/simd_impl/distances_arm_sve.cpp
 - vendor/faiss/faiss/utils/simd_impl/distances_autovec-inl.h
 - vendor/faiss/faiss/utils/simd_impl/distances_avx2.cpp
 - vendor/faiss/faiss/utils/simd_impl/distances_avx512.cpp
+- vendor/faiss/faiss/utils/simd_impl/distances_rvv.cpp
+- vendor/faiss/faiss/utils/simd_impl/distances_simdlib256.h
 - vendor/faiss/faiss/utils/simd_impl/distances_sse-inl.h
+- vendor/faiss/faiss/utils/simd_impl/exhaustive_L2sqr_blas_cmax.h
+- vendor/faiss/faiss/utils/simd_impl/hamming_impl.h
+- vendor/faiss/faiss/utils/simd_impl/partitioning_avx2.cpp
+- vendor/faiss/faiss/utils/simd_impl/partitioning_neon.cpp
+- vendor/faiss/faiss/utils/simd_impl/partitioning_simdlib256.h
+- vendor/faiss/faiss/utils/simd_impl/rabitq_avx2.cpp
+- vendor/faiss/faiss/utils/simd_impl/rabitq_avx512.cpp
+- vendor/faiss/faiss/utils/simd_impl/rabitq_neon.cpp
+- vendor/faiss/faiss/utils/simd_impl/rabitq_rvv.cpp
+- vendor/faiss/faiss/utils/simd_impl/super_kmeans_dispatch.h
+- vendor/faiss/faiss/utils/simd_impl/super_kmeans_kernels.h
+- vendor/faiss/faiss/utils/simd_impl/super_kmeans_kernels_avx2.cpp
+- vendor/faiss/faiss/utils/simd_impl/super_kmeans_kernels_avx512.cpp
 - vendor/faiss/faiss/utils/simd_levels.cpp
 - vendor/faiss/faiss/utils/simd_levels.h
-- vendor/faiss/faiss/utils/simdlib.h
-- vendor/faiss/faiss/utils/simdlib_avx2.h
-- vendor/faiss/faiss/utils/simdlib_avx512.h
-- vendor/faiss/faiss/utils/simdlib_emulated.h
-- vendor/faiss/faiss/utils/simdlib_neon.h
-- vendor/faiss/faiss/utils/simdlib_ppc64.h
 - vendor/faiss/faiss/utils/sorting.cpp
 - vendor/faiss/faiss/utils/sorting.h
 - vendor/faiss/faiss/utils/transpose/transpose-avx2-inl.h
@@ -436,7 +521,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 4.0.6
+rubygems_version: 4.0.10
 specification_version: 4
 summary: Efficient similarity search and clustering for Ruby
 test_files: []

data/vendor/faiss/faiss/impl/RaBitQStats.cpp

@@ -1,29 +0,0 @@
-/*
- * Copyright (c) Meta Platforms, Inc. and affiliates.
- *
- * This source code is licensed under the MIT license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-#include <faiss/impl/RaBitQStats.h>
-
-namespace faiss {
-
-// NOLINTNEXTLINE(facebook-avoid-non-const-global-variables)
-RaBitQStats rabitq_stats;
-
-void RaBitQStats::reset() {
-    n_1bit_evaluations = 0;
-    n_multibit_evaluations = 0;
-}
-
-double RaBitQStats::skip_percentage() const {
-    const size_t copy_n_1bit_evaluations = n_1bit_evaluations;
-    const size_t copy_n_multibit_evaluations = n_multibit_evaluations;
-    return copy_n_1bit_evaluations > 0
-            ? 100.0 * (copy_n_1bit_evaluations - copy_n_multibit_evaluations) /
-                    copy_n_1bit_evaluations
-            : 0.0;
-}
-
-} // namespace faiss

data/vendor/faiss/faiss/impl/RaBitQStats.h

@@ -1,56 +0,0 @@
-/*
- * Copyright (c) Meta Platforms, Inc. and affiliates.
- *
- * This source code is licensed under the MIT license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-#pragma once
-
-#include <faiss/impl/platform_macros.h>
-#include <cstddef>
-
-namespace faiss {
-
-/// Statistics for RaBitQ multi-bit two-stage search.
-///
-/// These stats are ONLY collected for multi-bit mode (nb_bits > 1).
-/// In 1-bit mode, there is no two-stage filtering - all candidates are
-/// evaluated with a single distance computation, so there is nothing
-/// meaningful to track. For 1-bit mode, both counters remain 0.
-///
-/// Multi-bit mode uses a two-stage search:
-///   Stage 1: Compute 1-bit lower bound distance for all candidates
-///   Stage 2: Compute full multi-bit distance only for promising candidates
-///
-/// The skip_percentage() metric measures filtering effectiveness:
-/// how many candidates were filtered out by the 1-bit lower bound
-/// without needing the more expensive multi-bit distance computation.
-///
-/// WARNING: Statistics are not robust to internal threading nor to
-/// concurrent RaBitQ searches. Use these values in a single-threaded
-/// context to accurately gauge RaBitQ's filtering effectiveness.
-/// Call reset() before search, then read stats after search completes.
-struct RaBitQStats {
-    /// Number of candidates evaluated using 1-bit (lower bound) distance.
-    /// This is the first stage of two-stage search in multi-bit mode.
-    /// Always 0 in 1-bit mode (stats not tracked).
-    size_t n_1bit_evaluations = 0;
-
-    /// Number of candidates that passed 1-bit filtering and required
-    /// full multi-bit distance computation (second stage).
-    /// Always 0 in 1-bit mode (stats not tracked).
-    size_t n_multibit_evaluations = 0;
-
-    void reset();
-
-    /// Compute percentage of candidates skipped (filtered out by 1-bit stage).
-    /// Returns 0 if no candidates were evaluated (including 1-bit mode).
-    double skip_percentage() const;
-};
-
-/// Global stats for RaBitQ indexes
-// NOLINTNEXTLINE(facebook-avoid-non-const-global-variables)
-FAISS_API extern RaBitQStats rabitq_stats;
-
-} // namespace faiss

data/vendor/faiss/faiss/impl/pq4_fast_scan.h

@@ -1,224 +0,0 @@
-/*
- * Copyright (c) Meta Platforms, Inc. and affiliates.
- *
- * This source code is licensed under the MIT license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-#pragma once
-
-#include <cstdint>
-#include <cstdlib>
-
-#include <faiss/impl/CodePacker.h>
-
-/** PQ4 SIMD packing and accumulation functions
- *
- * The basic kernel accumulates nq query vectors with bbs = nb * 2 * 16 vectors
- * and produces an output matrix for that. It is interesting for nq * nb <= 4,
- * otherwise register spilling becomes too large.
- *
- * The implementation of these functions is spread over 3 cpp files to reduce
- * parallel compile times. Templates are instantiated explicitly.
- */
-
-namespace faiss {
-
-struct NormTableScaler;
-struct SIMDResultHandler;
-
-/** Pack codes for consumption by the SIMD kernels.
- *  The unused bytes are set to 0.
- *
- * @param codes   input codes, size (ntotal, ceil(M / 2))
- * @param ntotal  number of input codes
- * @param nb      output number of codes (ntotal rounded up to a multiple of
- *                bbs)
- * @param nsq      number of sub-quantizers (=M rounded up to a multiple of 2)
- * @param bbs     size of database blocks (multiple of 32)
- * @param blocks  output array, size nb * nsq / 2.
- * @param code_stride  optional stride between consecutive codes (0 = use
-default (M + 1) / 2)
- */
-void pq4_pack_codes(
-        const uint8_t* codes,
-        size_t ntotal,
-        size_t M,
-        size_t nb,
-        size_t bbs,
-        size_t nsq,
-        uint8_t* blocks,
-        size_t code_stride = 0);
-
-/** Same as pack_codes but write in a given range of the output,
- * leaving the rest untouched. Assumes allocated entries are 0 on input.
- *
- * @param codes   input codes, size (i1 - i0, ceil(M / 2))
- * @param i0      first output code to write
- * @param i1      last output code to write
- * @param blocks  output array, size at least ceil(i1 / bbs) * bbs * nsq / 2
- * @param code_stride  optional stride between consecutive codes (0 = use
- * default (M + 1) / 2)
- * @param block_stride  stride in bytes between consecutive blocks.
- */
-void pq4_pack_codes_range(
-        const uint8_t* codes,
-        size_t M,
-        size_t i0,
-        size_t i1,
-        size_t bbs,
-        size_t nsq,
-        uint8_t* blocks,
-        size_t code_stride,
-        size_t block_stride);
-
-/** get a single element from a packed codes table
- *
- * @param vector_id        vector id
- * @param sq       subquantizer (< nsq)
- */
-uint8_t pq4_get_packed_element(
-        const uint8_t* data,
-        size_t bbs,
-        size_t nsq,
-        size_t vector_id,
-        size_t sq);
-
-/** set a single element "code" into a packed codes table
- *
- * @param vector_id       vector id
- * @param sq       subquantizer (< nsq)
- */
-void pq4_set_packed_element(
-        uint8_t* data,
-        uint8_t code,
-        size_t bbs,
-        size_t nsq,
-        size_t vector_id,
-        size_t sq);
-
-/** CodePacker API for the PQ4 fast-scan */
-struct CodePackerPQ4 : CodePacker {
-    size_t nsq;
-
-    CodePackerPQ4(size_t nsq, size_t bbs);
-
-    CodePacker* clone() const final;
-
-    void pack_1(const uint8_t* flat_code, size_t offset, uint8_t* block)
-            const final;
-    void unpack_1(const uint8_t* block, size_t offset, uint8_t* flat_code)
-            const final;
-};
-
-/** Pack Look-up table for consumption by the kernel.
- *
- * @param nq      number of queries
- * @param nsq     number of sub-quantizers (multiple of 2)
- * @param src     input array, size (nq, 16)
- * @param dest    output array, size (nq, 16)
- */
-void pq4_pack_LUT(int nq, int nsq, const uint8_t* src, uint8_t* dest);
-
-/** Loop over database elements and accumulate results into result handler
- *
- * @param nq      number of queries
- * @param nb      number of database elements
- * @param bbs     size of database blocks (multiple of 32)
- * @param nsq     number of sub-quantizers (multiple of 2)
- * @param codes   packed codes array
- * @param LUT     packed look-up table
- * @param scaler  scaler to scale the encoded norm
- * @param block_stride  stride in bytes between consecutive blocks.
- */
-void pq4_accumulate_loop(
-        int nq,
-        size_t nb,
-        int bbs,
-        int nsq,
-        const uint8_t* codes,
-        const uint8_t* LUT,
-        SIMDResultHandler& res,
-        const NormTableScaler* scaler,
-        size_t block_stride);
-
-/* qbs versions, supported only for bbs=32.
- *
- * The kernel function runs the kernel for *several* query blocks
- * and bbs database vectors. The sizes of the blocks are encoded in qbs as
- * base-16 digits.
- *
- * For example, qbs = 0x1223 means that the kernel will be run 4 times, the
- * first time with 3 query vectors, second time with 2 query vectors, then 2
- * vectors again and finally with 1 query vector. The output block will thus be
- * nq = 3 + 2 + 2 + 1 = 6 queries. For a given total block size, the optimal
- * decomposition into sub-blocks (measured empirically) is given by
- * preferred_qbs().
- */
-
-/* compute the number of queries from a base-16 decomposition */
-int pq4_qbs_to_nq(int qbs);
-
-/** return the preferred decomposition in blocks for a nb of queries. */
-int pq4_preferred_qbs(int nq);
-
-/** Pack Look-up table for consumption by the kernel.
- *
- * @param qbs     4-bit encoded number of query blocks, the total number of
- *                queries handled (nq) is deduced from it
- * @param nsq     number of sub-quantizers (multiple of 2)
- * @param src     input array, size (nq, 16)
- * @param dest    output array, size (nq, 16)
- * @return nq
- */
-int pq4_pack_LUT_qbs(int fqbs, int nsq, const uint8_t* src, uint8_t* dest);
-
-/** Same as pq4_pack_LUT_qbs, except the source vectors are remapped with q_map
- */
-int pq4_pack_LUT_qbs_q_map(
-        int qbs,
-        int nsq,
-        const uint8_t* src,
-        const int* q_map,
-        uint8_t* dest);
-
-/** Run accumulation loop.
- *
- * @param qbs     4-bit encoded number of queries
- * @param nb      number of database codes (multiple of bbs)
- * @param nsq     number of sub-quantizers
- * @param codes   encoded database vectors (packed)
- * @param LUT     look-up table (packed)
- * @param res     call-back for the results
- * @param scaler  scaler to scale the encoded norm
- * @param block_stride  stride in bytes between consecutive blocks.
- */
-void pq4_accumulate_loop_qbs(
-        int qbs,
-        size_t nb,
-        int nsq,
-        const uint8_t* codes,
-        const uint8_t* LUT,
-        SIMDResultHandler& res,
-        const NormTableScaler* scaler,
-        size_t block_stride);
-
-/** Wrapper of pq4_accumulate_loop_qbs using simple StoreResultHandler
- *  and DummyScaler
- *
- * @param nq      number of queries
- * @param ntotal2 number of database elements (multiple of 32)
- * @param nsq     number of sub-quantizers (muliple of 2)
- * @param codes   packed codes array
- * @param LUT     packed look-up table
- * @param accu    array to store the results
- */
-void accumulate_to_mem(
-        int nq,
-        size_t ntotal2,
-        int nsq,
-        const uint8_t* codes,
-        const uint8_t* LUT,
-        uint16_t* accu);
-
-} // namespace faiss