diff --git a/bench/batch-matrix-multiply.cc b/bench/batch-matrix-multiply.cc
index b09cdebc34b..c2affbc14e3 100644
--- a/bench/batch-matrix-multiply.cc
+++ b/bench/batch-matrix-multiply.cc
@@ -28,8 +28,10 @@
 void xnnpack_batch_matrix_multiply_f32(benchmark::State& state, const char* net) {
   const size_t batch_size = state.range(0);
   const size_t m = state.range(1);
-  const size_t k = state.range(1);
-  const size_t n = state.range(1);
+  const size_t k = state.range(2);
+  const size_t n = state.range(3);
+  const bool transpose_b = state.range(4);
+  const size_t num_threads = state.range(5);
 
   std::random_device random_device;
   auto rng = std::mt19937(random_device());
@@ -53,8 +55,9 @@ void xnnpack_batch_matrix_multiply_f32(benchmark::State& state, const char* net)
 
   xnnpack::Buffer<xnn_operator_t> ops(num_buffers);
 
+  uint32_t flags = transpose_b ? XNN_FLAG_TRANSPOSE_B : 0;
   for (xnn_operator_t& op : ops) {
-    status = xnn_create_batch_matrix_multiply_nc_f32(/*flags=*/0, &op);
+    status = xnn_create_batch_matrix_multiply_nc_f32(flags, &op);
     if (status != xnn_status_success) {
       state.SkipWithError("failed to create FP32 Convolution operator");
       return;
@@ -85,12 +88,14 @@ void xnnpack_batch_matrix_multiply_f32(benchmark::State& state, const char* net)
   }
 
   size_t buffer_index = 0;
+  pthreadpool_t threadpool = pthreadpool_create(num_threads);
+
   for (auto _ : state) {
     state.PauseTiming();
     buffer_index = (buffer_index + 1) % num_buffers;
     state.ResumeTiming();
 
-    status = xnn_run_operator(ops[buffer_index], /*threadpool=*/nullptr);
+    status = xnn_run_operator(ops[buffer_index], threadpool);
     if (status != xnn_status_success) {
       state.SkipWithError("failed to run FP32 Convolution operator");
       return;
@@ -114,6 +119,8 @@ void xnnpack_batch_matrix_multiply_f32(benchmark::State& state, const char* net)
   state.counters["FLOPS"] = benchmark::Counter(
     uint64_t(state.iterations()) * batch_size * m * k * n,
     benchmark::Counter::kIsRate);
+
+  pthreadpool_destroy(threadpool);
 }
 
 #ifdef BENCHMARK_TENSORFLOW_LITE