[libcxx-commits] [libcxx] [libc++] Make the associative container query benchmarks more representative (PR #183036)

[Louis Dionne via libcxx-commits]

================
@@ -687,74 +687,32 @@ void associative_container_benchmarks(std::string container) {
   /////////////////////////
   // Query
   /////////////////////////
-  auto with_existent_key = [=](auto func) {
+  auto query_bench = [=](auto func) {
     return [=](auto& st) {
       const std::size_t size = st.range(0);
       std::vector<Value> in  = make_value_types(generate_unique_keys(size));
-      // Pick any `BatchSize` number of elements
-      std::vector<Key> keys;
-      for (std::size_t i = 0; i < in.size(); i += (in.size() / BatchSize)) {
-        keys.push_back(get_key(in.at(i)));
-      }
       Container c(in.begin(), in.end());
 
-      while (st.KeepRunningBatch(BatchSize)) {
-        for (std::size_t i = 0; i != keys.size(); ++i) { // possible empty keys when Arg(0)
-          auto result = func(c, keys[i]);
-          benchmark::DoNotOptimize(c);
-          benchmark::DoNotOptimize(result);
-          benchmark::ClobberMemory();
-        }
-      }
-    };
-  };
-
-  auto with_nonexistent_key = [=](auto func) {
----------------
ldionne wrote:

As discussed, let's try implementing the non-existent benchmark in a way that reduces branch predictions. Then, if the existent and non-existent cases have basically the same execution time, I think it would be fine to remove the non-existent case.

https://github.com/llvm/llvm-project/pull/183036