diff --git a/faiss/perf_tests/bench_hnsw_add.cpp b/faiss/perf_tests/bench_hnsw_add.cpp
new file mode 100644
index 0000000000..f975360803
--- /dev/null
+++ b/faiss/perf_tests/bench_hnsw_add.cpp
@@ -0,0 +1,74 @@
+/**
+ * Copyright (c) Facebook, Inc. and its affiliates.
+ *
+ * This source code is licensed under the MIT license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+#include <fmt/format.h>
+#include <gflags/gflags.h>
+#include <omp.h>
+
+#include <benchmark/benchmark.h>
+#include <faiss/perf_tests/utils.h>
+#include <faiss/utils/distances.h>
+#include <faiss/utils/random.h>
+#include <faiss/utils/utils.h>
+#include "faiss/IndexHNSW.h"
+
+using namespace faiss;
+DEFINE_uint32(d, 128, "dimension");
+DEFINE_uint32(n, 2000, "dimension");
+DEFINE_uint32(iterations, 20, "iterations");
+DEFINE_uint32(M, 32, "connections per vertex");
+DEFINE_uint32(num_omp_threads, 1, "number of omp threads");
+
+DEFINE_uint32(
+        efConstruction,
+        40,
+        "size of the dynamic list containing the nearest neighbors, which is used during index time");
+
+static void bench_add(
+        benchmark::State& state,
+        int num_threads,
+        int d,
+        int n,
+        int M,
+        int efConstruction) {
+    std::vector<float> x(d * n);
+
+    float_rand(x.data(), d * n, 12345);
+    faiss::IndexHNSWFlat index(d, M);
+    index.hnsw.efConstruction = efConstruction;
+
+    omp_set_num_threads(num_threads);
+
+    for (auto _ : state) {
+        index.add(n, x.data());
+    }
+}
+
+int main(int argc, char** argv) {
+    benchmark::Initialize(&argc, argv);
+    gflags::AllowCommandLineReparsing();
+    gflags::ParseCommandLineFlags(&argc, &argv, true);
+
+    int iterations = FLAGS_iterations;
+    int d = FLAGS_d;
+    int n = FLAGS_n;
+    int M = FLAGS_M;
+    int efConstruction = FLAGS_efConstruction;
+    int num_threads = FLAGS_num_omp_threads;
+    benchmark::RegisterBenchmark(
+            fmt::format("hnsw_add_{}d_{}n_{}M", d, n, M).c_str(),
+            bench_add,
+            num_threads,
+            d,
+            n,
+            M,
+            efConstruction)
+            ->Iterations(iterations);
+
+    benchmark::RunSpecifiedBenchmarks();
+    benchmark::Shutdown();
+}
diff --git a/faiss/perf_tests/bench_scalar_quantizer_accuracy.cpp b/faiss/perf_tests/bench_scalar_quantizer_accuracy.cpp
index 7d8e70f958..4b7ef48a53 100644
--- a/faiss/perf_tests/bench_scalar_quantizer_accuracy.cpp
+++ b/faiss/perf_tests/bench_scalar_quantizer_accuracy.cpp
@@ -27,7 +27,6 @@ static void bench_reconstruction_error(
         ScalarQuantizer::QuantizerType type,
         int d,
         int n) {
-    state.SetLabel(faiss::get_compile_options());
     std::vector<float> x(d * n);
 
     float_rand(x.data(), d * n, 12345);
@@ -64,7 +63,6 @@ static void bench_reconstruction_error(
     state.counters["ndiff_for_idempotence"] = ndiff;
 
     state.counters["code_size_two"] = codes.size();
-    state.SetLabel(faiss::get_compile_options());
 }
 
 int main(int argc, char** argv) {
diff --git a/faiss/perf_tests/bench_scalar_quantizer_decode.cpp b/faiss/perf_tests/bench_scalar_quantizer_decode.cpp
index 6cf3713a9b..c2fdbf4294 100644
--- a/faiss/perf_tests/bench_scalar_quantizer_decode.cpp
+++ b/faiss/perf_tests/bench_scalar_quantizer_decode.cpp
@@ -27,7 +27,6 @@ static void bench_decode(
         ScalarQuantizer::QuantizerType type,
         int d,
         int n) {
-    state.SetLabel(faiss::get_compile_options());
     std::vector<float> x(d * n);
 
     float_rand(x.data(), d * n, 12345);
diff --git a/faiss/perf_tests/bench_scalar_quantizer_distance.cpp b/faiss/perf_tests/bench_scalar_quantizer_distance.cpp
index b621ed39b3..81a92d7c7b 100644
--- a/faiss/perf_tests/bench_scalar_quantizer_distance.cpp
+++ b/faiss/perf_tests/bench_scalar_quantizer_distance.cpp
@@ -27,7 +27,6 @@ static void bench_distance(
         ScalarQuantizer::QuantizerType type,
         int n,
         int d) {
-    state.SetLabel(faiss::get_compile_options());
     std::vector<float> x(d * n);
 
     float_rand(x.data(), d * n, 12345);
@@ -46,8 +45,6 @@ static void bench_distance(
     std::vector<uint8_t> codes(code_size * n);
     sq.compute_codes(x.data(), codes.data(), n);
 
-    state.SetLabel(faiss::get_compile_options());
-
     std::unique_ptr<ScalarQuantizer::SQDistanceComputer> dc(
             sq.get_distance_computer());
     dc->codes = codes.data();
diff --git a/faiss/perf_tests/bench_scalar_quantizer_encode.cpp b/faiss/perf_tests/bench_scalar_quantizer_encode.cpp
index 3f745ada9a..c35ecbd58c 100644
--- a/faiss/perf_tests/bench_scalar_quantizer_encode.cpp
+++ b/faiss/perf_tests/bench_scalar_quantizer_encode.cpp
@@ -28,7 +28,6 @@ static void bench_encode(
         ScalarQuantizer::QuantizerType type,
         int d,
         int n) {
-    state.SetLabel(faiss::get_compile_options());
     std::vector<float> x(d * n);
 
     float_rand(x.data(), d * n, 12345);