From 45f9ea1e798c2e128dcba53602ccbddfe5676f6f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Martin=20Ho=C5=99e=C5=88ovsk=C3=BD?=
 <martin.horenovsky@gmail.com>
Date: Sat, 22 Aug 2026 15:19:33 +0200
Subject: [PATCH] Add optimizer barrier to benchmark calls of void-returning
 functions

The goal is to do the equivalent of clobbering memory, which forces
the compiler to keep side effects that are external to the benchmarked
function, while allowing it to optimize inside the benchmarked function.

E.g. this cannot be optimized away:
```cpp
BENCHMARK("foo") {
    global_count += 1;
};
```

while this can:
```cpp
BENCHMARK("bar") {
    size_t local_count = 0;
    for (size_t i = 0; i < 100; ++i) {
        local_count += 1;
    }
};
```
---
 src/catch2/benchmark/catch_optimizer.hpp            |  9 ++++++++-
 tests/CMakeLists.txt                                | 13 -------------
 .../IntrospectiveTests/InternalBenchmark.tests.cpp  |  3 ---
 3 files changed, 8 insertions(+), 17 deletions(-)

diff --git a/src/catch2/benchmark/catch_optimizer.hpp b/src/catch2/benchmark/catch_optimizer.hpp
index 61e6571f..5535b2a2 100644
--- a/src/catch2/benchmark/catch_optimizer.hpp
+++ b/src/catch2/benchmark/catch_optimizer.hpp
@@ -34,7 +34,7 @@ namespace Catch {
         } // namespace Detail
 #elif defined(_MSC_VER) || defined(__IAR_SYSTEMS_ICC__)
 
-#if defined(_MSVC_VER)
+#if defined(_MSC_VER)
 #pragma optimize("", off)
 #elif defined(__IAR_SYSTEMS_ICC__)
 // For IAR the pragma only affects the following function
@@ -71,6 +71,13 @@ namespace Catch {
         template <typename Fn, typename... Args>
         inline auto invoke_deoptimized(Fn&& fn, Args&&... args) -> std::enable_if_t<std::is_same<void, decltype(fn(args...))>::value> {
             CATCH_FORWARD((fn)) (CATCH_FORWARD(args)...);
+            // In the non-void case, we pass the result through `deoptimize_value`
+            // to force the compiler to keep it. We have no return value here,
+            // but add an optimizer barrier (ideally a memory clobber) to force
+            // the _side effects_ of the loop be visible (e.g. writes to globals).
+            // Note that writes to benchmark-locals can be optimized away, as
+            // we would expect in normal code.
+            Detail::optimizer_barrier();
         }
     } // namespace Benchmark
 } // namespace Catch
diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt
index 5a74a9d0..469e3f3d 100644
--- a/tests/CMakeLists.txt
+++ b/tests/CMakeLists.txt
@@ -486,19 +486,6 @@ set_tests_properties("Benchmarking::SkipBenchmarkMacros"
     FAIL_REGULAR_EXPRESSION "benchmark name"
 )
 
-add_test(NAME "Benchmarking::FailureReporting::OptimizedOut"
-  COMMAND
-    $<TARGET_FILE:SelfTest> "Failing benchmarks" -c "empty" -r xml
-  # This test only makes sense with the optimizer being enabled when
-  # the tests are being compiled.
-  CONFIGURATIONS Release
-)
-set_tests_properties("Benchmarking::FailureReporting::OptimizedOut"
-  PROPERTIES
-    PASS_REGULAR_EXPRESSION "could not measure benchmark\, maybe it was optimized away"
-    FAIL_REGULAR_EXPRESSION "successes=\"1\""
-)
-
 add_test(NAME "Benchmarking::FailureReporting::ThrowingBenchmark"
   COMMAND
     $<TARGET_FILE:SelfTest> "Failing benchmarks" -c "throw" -r xml
diff --git a/tests/SelfTest/IntrospectiveTests/InternalBenchmark.tests.cpp b/tests/SelfTest/IntrospectiveTests/InternalBenchmark.tests.cpp
index 69251d97..b7ec17b8 100644
--- a/tests/SelfTest/IntrospectiveTests/InternalBenchmark.tests.cpp
+++ b/tests/SelfTest/IntrospectiveTests/InternalBenchmark.tests.cpp
@@ -429,9 +429,6 @@ TEST_CASE("run benchmark", "[benchmark][approvals]") {
 }
 
 TEST_CASE("Failing benchmarks", "[!benchmark][.approvals]") {
-    SECTION("empty", "Benchmark that has been optimized away (because it is empty)") {
-        BENCHMARK("Empty benchmark") {};
-    }
     SECTION("throw", "Benchmark that throws an exception") {
         BENCHMARK("Throwing benchmark") {
             throw "just a plain literal, bleh";