Fix implicit integer precision loss from 64-bit to 32-bit (#5091)

Summary:
Pull Request resolved: #5091

Fix 172 `clang-diagnostic-shorten-64-to-32` lint warnings across 29 files by adding explicit `static_cast<int>()` or widening variable types where `size_t`/`idx_t` (64-bit) values were implicitly narrowed to `int`/`int32_t` (32-bit).

The fixes fall into two categories:
- **Explicit casts**: Where the receiving API requires `int` and the value is known to fit (e.g., vector dimensions, sub-quantizer counts, cluster counts, BLAS parameters)
- **Type widening**: Where the variable was unnecessarily narrow (e.g., `int nprobe` → `size_t nprobe`, `int list_no` → `size_t list_no`)

Differential Revision: D100588996

Author: junjieqi

faiss/AutoTune.cpp

@@ -174,7 +174,7 @@ double OperatingPoints::t_for_perf(double perf) const {
     if (perf > a.back().perf) {
         return 1e50;
     }
-    int i0 = -1, i1 = a.size() - 1;
+    int i0 = -1, i1 = static_cast<int>(a.size()) - 1;
     while (i0 + 1 < i1) {
         int imed = (i0 + i1 + 1) / 2;
         if (a[imed].perf < perf) {
@@ -786,7 +786,7 @@ void ParameterSpace::explore(
     // make sure the slowest and fastest experiment are run
     perm[0] = 0;
     if (n_comb > 1) {
-        perm[1] = n_comb - 1;
+        perm[1] = static_cast<int>(n_comb - 1);
         rand_perm(&perm[2], n_comb - 2, 1234);
         for (size_t i = 2; i < perm.size(); i++) {
             perm[i]++;

faiss/Clustering.cpp

@@ -92,7 +92,7 @@ idx_t subsample_training_set(
         const idx_t new_nx = clus.k * clus.max_points_per_centroid;
         perm.resize(new_nx);
         for (idx_t i = 0; i < new_nx; i++) {
-            perm[i] = rng.rand_int(nx);
+            perm[i] = rng.rand_int(static_cast<int>(nx));
         }
     } else {
         // use subsampling with a default std rng
@@ -262,7 +262,7 @@ int split_clusters(
         }
     }
 
-    return nsplit;
+    return static_cast<int>(nsplit);
 }
 
 } // namespace
@@ -538,7 +538,7 @@ void Clustering::train_encoded(
                     obj,
                     (getmillisecs() - t0) / 1000.0,
                     t_search_tot / 1000,
-                    imbalance_factor(nx, k, assign.get()),
+                    imbalance_factor(nx, static_cast<int>(k), assign.get()),
                     nsplit};
             iteration_stats.push_back(stats);
 
@@ -649,7 +649,7 @@ float kmeans_clustering(
         size_t k,
         const float* x,
         float* centroids) {
-    Clustering clus(d, k);
+    Clustering clus(static_cast<int>(d), static_cast<int>(k));
     clus.verbose = d * n * k > (size_t(1) << 30);
     // display logs if > 1Gflop per iteration
     IndexFlatL2 index(d);
@@ -700,7 +700,7 @@ void ProgressiveDimClustering::train(
         ProgressiveDimIndexFactory& factory) {
     int d_prev = 0;
 
-    PCAMatrix pca(d, d);
+    PCAMatrix pca(static_cast<int>(d), static_cast<int>(d));
 
     std::vector<float> xbuf;
     if (apply_pca) {
@@ -725,7 +725,7 @@ void ProgressiveDimClustering::train(
         }
         std::unique_ptr<Index> clustering_index(factory(di));
 
-        Clustering clus(di, k, *this);
+        Clustering clus(di, static_cast<int>(k), *this);
         if (d_prev > 0) {
             // copy warm-start centroids (padded with 0s)
             clus.centroids.resize(k * di);

faiss/IVFlib.cpp

@@ -465,7 +465,7 @@ IndexIVFResidualQuantizer* ivf_residual_from_quantizer(
     std::vector<size_t> nbits(nlevel);
     std::copy(rq.nbits.begin(), rq.nbits.begin() + nlevel, nbits.begin());
     std::unique_ptr<ResidualCoarseQuantizer> rcq(
-            new ResidualCoarseQuantizer(rq.d, nbits));
+            new ResidualCoarseQuantizer(static_cast<int>(rq.d), nbits));
 
     // set the coarse quantizer from the 2 first quantizers
     rcq->rq.initialize_from(rq);
@@ -528,22 +528,24 @@ void ivf_residual_add_from_flat_codes(
             for (idx_t i = 0; i < static_cast<idx_t>(nb); i++) {
                 const uint8_t* code = &raw_codes[i * code_size];
                 BitstringReader rd(code, code_size);
-                idx_t list_no = rd.read(rcq->rq.tot_bits);
+                idx_t list_no = rd.read(static_cast<int>(rcq->rq.tot_bits));
 
                 if (list_no % nt ==
                     rank) { // each thread takes care of 1/nt of the invlists
                     // copy AQ indexes one by one
                     BitstringWriter wr(tmp_code.data(), tmp_code.size());
                     for (size_t j = 0; j < rq.M; j++) {
-                        int nbit = rq.nbits[j];
+                        int nbit = static_cast<int>(rq.nbits[j]);
                         wr.write(rd.read(nbit), nbit);
                     }
                     // we need to recompute the norm
                     // decode first, does not use the norm component, so that's
                     // ok
                     index->rq.decode(tmp_code.data(), tmp.data(), 1);
                     float norm = fvec_norm_L2sqr<SL>(tmp.data(), rq.d);
-                    wr.write(rq.encode_norm(norm), rq.norm_bits);
+                    wr.write(
+                            rq.encode_norm(norm),
+                            static_cast<int>(rq.norm_bits));
 
                     // add code to the inverted list
                     invlists.add_entry(list_no, i, tmp_code.data());

faiss/Index.h

@@ -115,7 +115,7 @@ struct Index {
     float metric_arg; ///< argument of the metric type
 
     explicit Index(idx_t d_in = 0, MetricType metric = METRIC_L2)
-            : d(d_in),
+            : d(static_cast<int>(d_in)),
               ntotal(0),
               verbose(false),
               is_trained(true),

faiss/Index2Layer.cpp

@@ -171,7 +171,7 @@ struct DistanceXPQ4 : Distance2Level {
                 dynamic_cast<IndexFlat*>(storage.q1.quantizer);
 
         FAISS_ASSERT(quantizer);
-        M = storage.pq.M;
+        M = static_cast<int>(storage.pq.M);
         pq_l1_tab = quantizer->get_xb();
     }
 
@@ -218,8 +218,8 @@ struct Distance2xXPQ4 : Distance2Level {
 
         FAISS_ASSERT(mi);
         FAISS_ASSERT(storage.pq.M % 2 == 0);
-        M_2 = storage.pq.M / 2;
-        mi_nbits = mi->pq.nbits;
+        M_2 = static_cast<int>(storage.pq.M / 2);
+        mi_nbits = static_cast<int>(mi->pq.nbits);
         pq_l1_tab = mi->pq.centroids.data();
     }
 

faiss/IndexAdditiveQuantizer.cpp

@@ -543,7 +543,7 @@ void ResidualCoarseQuantizer::search(
 
     int beam_size = int(k * actual_beam_factor);
     if (beam_size > ntotal) {
-        beam_size = ntotal;
+        beam_size = static_cast<int>(ntotal);
     }
     size_t memory_per_point = rq.memory_per_point(beam_size);
 

faiss/IndexAdditiveQuantizerFastScan.cpp

@@ -56,7 +56,7 @@ void IndexAdditiveQuantizerFastScan::init(
     } else {
         M = aq_init->M;
     }
-    init_fastscan(aq_init->d, M, 4, metric, bbs_);
+    init_fastscan(static_cast<int>(aq_init->d), M, 4, metric, bbs_);
 
     max_train_points = 1024 * ksub * M;
 }

faiss/IndexBinary.cpp

@@ -17,7 +17,9 @@
 namespace faiss {
 
 IndexBinary::IndexBinary(idx_t d_, MetricType metric)
-        : d(d_), code_size(d_ / 8), metric_type(metric) {
+        : d(static_cast<int>(d_)),
+          code_size(static_cast<int>(d_ / 8)),
+          metric_type(metric) {
     FAISS_THROW_IF_NOT(d_ % 8 == 0);
 }
 

faiss/IndexBinaryHNSW.cpp

@@ -74,7 +74,8 @@ void hnsw_add_vertices(
 
         // build histogram
         for (size_t i = 0; i < n; i++) {
-            HNSW::storage_idx_t pt_id = i + n0;
+            HNSW::storage_idx_t pt_id =
+                    static_cast<HNSW::storage_idx_t>(i + n0);
             int pt_level = hnsw.levels[pt_id] - 1;
             while (pt_level >= static_cast<int>(hist.size())) {
                 hist.push_back(0);
@@ -90,7 +91,8 @@ void hnsw_add_vertices(
 
         // bucket sort
         for (size_t i = 0; i < n; i++) {
-            HNSW::storage_idx_t pt_id = i + n0;
+            HNSW::storage_idx_t pt_id =
+                    static_cast<HNSW::storage_idx_t>(i + n0);
             int pt_level = hnsw.levels[pt_id] - 1;
             order[offsets[pt_level]++] = pt_id;
         }
@@ -99,7 +101,7 @@ void hnsw_add_vertices(
     { // perform add
         RandomGenerator rng2(789);
 
-        int i1 = n;
+        int i1 = static_cast<int>(n);
 
         for (int pt_level = static_cast<int>(hist.size()) - 1;
              pt_level >= int(!index_hnsw.init_level0);
@@ -257,7 +259,7 @@ void IndexBinaryHNSW::search(
 
 void IndexBinaryHNSW::add(idx_t n, const uint8_t* x) {
     FAISS_THROW_IF_NOT(is_trained);
-    int n0 = ntotal;
+    idx_t n0 = ntotal;
     storage->add(n, x);
     ntotal = storage->ntotal;
 
@@ -390,7 +392,7 @@ void IndexBinaryHNSWCagra::search(
                 float distance = (*dis)(idx);
 
                 if (distance < nearest_d[i]) {
-                    nearest[i] = idx;
+                    nearest[i] = static_cast<storage_idx_t>(idx);
                     nearest_d[i] = distance;
                 }
             }

faiss/IndexBinaryIVF.cpp

@@ -151,7 +151,7 @@ void IndexBinaryIVF::search(
     indexIVF_stats.quantization_time += getmillisecs() - t0;
 
     t0 = getmillisecs();
-    invlists->prefetch_lists(idx.get(), n * nprobe_);
+    invlists->prefetch_lists(idx.get(), static_cast<int>(n * nprobe_));
 
     search_preassigned(
             n,
@@ -209,7 +209,7 @@ void IndexBinaryIVF::search_and_reconstruct(
 
     quantizer->search(n, x, nprobe_2, coarse_dis.get(), idx.get());
 
-    invlists->prefetch_lists(idx.get(), n * nprobe_2);
+    invlists->prefetch_lists(idx.get(), static_cast<int>(n * nprobe_2));
 
     // search_preassigned() with `store_pairs` enabled to obtain the list_no
     // and offset into `codes` for reconstruction
@@ -277,7 +277,7 @@ void IndexBinaryIVF::train(idx_t n, const uint8_t* x) {
             printf("Training quantizer on %" PRId64 " vectors in %dD\n", n, d);
         }
 
-        Clustering clus(d, nlist, cp);
+        Clustering clus(d, static_cast<int>(nlist), cp);
         quantizer->reset();
 
         IndexFlatL2 index_tmp(d);
@@ -705,9 +705,10 @@ void search_knn_hamming_per_invlist(
     int32_t* keys = new int32_t[n * nprobe];
     std::unique_ptr<int32_t[]> delete_keys(keys);
     for (size_t i = 0; i < n * static_cast<size_t>(nprobe); i++) {
-        keys[i] = keys_in[i];
+        keys[i] = static_cast<int32_t>(keys_in[i]);
     }
-    matrix_bucket_sort_inplace(n, nprobe, keys, ivf->nlist, lims.data(), 0);
+    matrix_bucket_sort_inplace(
+            n, nprobe, keys, static_cast<int32_t>(ivf->nlist), lims.data(), 0);
 
     using C = CMax<int32_t, idx_t>;
     heap_heapify<C>(n * k, distances, labels);
@@ -754,7 +755,12 @@ void search_knn_hamming_per_invlist(
         } else {
             for (; i + BS <= nq; i += BS) {
                 BlockSearchVariableK<HammingComputer, BS> bc(
-                        code_size, k, x, keys + l0 + i, distances, labels);
+                        code_size,
+                        static_cast<int>(k),
+                        x,
+                        keys + l0 + i,
+                        distances,
+                        labels);
                 for (idx_t j = 0; j < nb; j++) {
                     bc.add_bcode(bcodes + j * code_size, ids[j]);
                 }
@@ -764,7 +770,8 @@ void search_knn_hamming_per_invlist(
         // leftovers
         for (; i < nq; i++) {
             idx_t qno = keys[l0 + i];
-            HammingComputer hc(x + qno * code_size, code_size);
+            HammingComputer hc(
+                    x + qno * code_size, static_cast<int>(code_size));
             idx_t* __restrict idxi = labels + qno * k;
             int32_t* __restrict simi = distances + qno * k;
             int32_t simi0 = simi[0];
@@ -868,7 +875,7 @@ void IndexBinaryIVF::range_search(
     indexIVF_stats.quantization_time += getmillisecs() - t0;
 
     t0 = getmillisecs();
-    invlists->prefetch_lists(idx.get(), n * nprobe_2);
+    invlists->prefetch_lists(idx.get(), static_cast<int>(n * nprobe_2));
 
     range_search_preassigned(n, x, radius, idx.get(), coarse_dis.get(), res);