Add simple benchmark for vector forward mode

benchmark/Simple.cpp

@@ -55,5 +55,38 @@ static void BM_ReverseModeSumExecute(benchmark::State &state) {
 }
 BENCHMARK(BM_ReverseModeSumExecute);
 
+// Benchmark computing gradient using vector forward mode via
+// CladFunction::execute.
+static void BM_VectorForwardModeSumExecute(benchmark::State &state) {
+  auto vm_grad = clad::differentiate<clad::opts::vector_mode>(sum, "p");
+  double inputs[] = {1, 2, 3, 4, 5};
+  double result[3] = {};
+  clad::array_ref<double> result_ref(result, 3);
+  unsigned long long sum = 0;
+  for (auto _ : state) {
+    vm_grad.execute(inputs,/*dim*/ 3, result_ref);
+    benchmark::DoNotOptimize(sum += result[0] + result[1] + result[2]);
+  }
+}
+BENCHMARK(BM_VectorForwardModeSumExecute);
+
+// Benchmark computing gradient using vector forward mode via
+// a forward declaration.
+inline void sum_dvec_0(double*, int, clad::array_ref<double>);
+static void BM_VectorForwardModeSumFwdDecl(benchmark::State &state) {
+  auto vm_grad = clad::differentiate<clad::opts::vector_mode>(sum, "p");
+  (void) vm_grad;
+  double inputs[] = {1, 2, 3, 4, 5};
+  double result[3] = {};
+  clad::array_ref<double> result_ref(result, 3);
+  unsigned long long sum = 0;
+  for (auto _ : state) {
+    sum_dvec_0(inputs,/*dim*/ 3, result_ref);
+    benchmark::DoNotOptimize(sum += result[0] + result[1] + result[2]);
+  }
+}
+BENCHMARK(BM_VectorForwardModeSumFwdDecl);
+
+
 // Define our main.
 BENCHMARK_MAIN();