<__msvc_int128.hpp>: optimize _Signed128::operator* with a branchless multiply

benchmarks/CMakeLists.txt

@@ -137,6 +137,7 @@ add_benchmark(sample src/sample.cpp)
 add_benchmark(search src/search.cpp)
 add_benchmark(search_n src/search_n.cpp)
 add_benchmark(shuffle src/shuffle.cpp)
+add_benchmark(signed128_mul src/signed128_mul.cpp)
 add_benchmark(std_copy src/std_copy.cpp)
 add_benchmark(sv_equal src/sv_equal.cpp)
 add_benchmark(swap_ranges src/swap_ranges.cpp)

benchmarks/src/signed128_mul.cpp

@@ -0,0 +1,75 @@
+// Copyright (c) Microsoft Corporation.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+#include <benchmark/benchmark.h>
+
+#include <__msvc_int128.hpp>
+#include <array>
+#include <cstdint>
+#include <random>
+#include <utility>
+#include <vector>
+
+using namespace std;
+
+_NODISCARD constexpr _Signed128 old_mul128(_Signed128 _Left, _Signed128 _Right) {
+    bool _negative = false;
+    _Left._Strip_negative(_negative);
+    _Right._Strip_negative(_negative);
+    _Signed128 _Result{_Base128::_Multiply(_Left, _Right)};
+    if (_negative) {
+        _Result = -_Result;
+    }
+    return _Result;
+}
+
+vector<pair<_Signed128, _Signed128>> init_test_set() {
+    vector<pair<_Signed128, _Signed128>> vec(1'000'000);
+    mt19937_64 mt64{};
+    for (auto& [val1, val2] : vec) {
+        val1._Word[0] = mt64();
+        val1._Word[1] = mt64();
+        val2._Word[0] = mt64();
+        val2._Word[1] = mt64();
+    }
+    return vec;
+}
+
+void bm_signed128_mul(benchmark::State& state) {
+    const auto vec = init_test_set();
+
+    auto it        = vec.begin();
+    _Signed128 res = 0;
+
+    for (auto _ : state) {
+        res = it->first * it->second;
+
+        benchmark::DoNotOptimize(res);
+        ++it;
+        if (it == vec.end()) {
+            it = vec.begin();
+        }
+    }
+}
+
+void bm_signed128_oldmul(benchmark::State& state) {
+    const auto vec = init_test_set();
+
+    auto it        = vec.begin();
+    _Signed128 res = 0;
+
+    for (auto _ : state) {
+        res = old_mul128(it->first, it->second);
+
+        benchmark::DoNotOptimize(res);
+        ++it;
+        if (it == vec.end()) {
+            it = vec.begin();
+        }
+    }
+}
+
+BENCHMARK(bm_signed128_mul);
+BENCHMARK(bm_signed128_oldmul);
+
+BENCHMARK_MAIN();

stl/inc/__msvc_int128.hpp

@@ -1370,14 +1370,7 @@ struct _Signed128 : _Base128 {
     }
 
     _NODISCARD friend constexpr _Signed128 operator*(_Signed128 _Left, _Signed128 _Right) noexcept {
-        bool _Negative = false;
-        _Left._Strip_negative(_Negative);
-        _Right._Strip_negative(_Negative);
-        _Signed128 _Result{_Base128::_Multiply(_Left, _Right)};
-        if (_Negative) {
-            _Result = -_Result;
-        }
-        return _Result;
+        return _Signed128{_Base128::_Multiply(_Left, _Right)};
     }
 
     _TEMPLATE_CLASS_INTEGRAL(_Ty)