Function Benchmark

vectorization/src/main/cpp/benchmark.cpp:50–70

* Benchmarks a given matrix multiply operation. * * The multiply is given here as a callback to try to keep Clang from folding, * unrolling, or inlining inconsistently across each benchmarked implementation. * We want Clang to do as much as possible to optimize *within* the multiply * function itself, but inconsistent optimization of the benchmark code itself * could skew results. * * @par

 * @return The average duration per call in nanoseconds.
 */
[[nodiscard, clang::noinline]] std::chrono::nanoseconds Benchmark(
    Vec4<>& position, Mat4<>& translation,
    std::function<Vec4<>(const Vec4<>&, const Mat4<>&)> func) {
  // TODO: Move to a unit test.
  auto test = func(position, translation);
  auto expected = Vec4{{20, 10, 10, 1}};
  CHECK_EQ(test, expected);

  auto begin = std::chrono::steady_clock::now();

  // This is another attempt to prevent Clang from optimizing the benchmark
  // harness inconsistently.
#pragma clang loop unroll(disable)
  for (auto i = 0U; i < kNumRuns; i++) {
    result = func(position, translation);
  }

  auto end = std::chrono::steady_clock::now();

  return (end - begin) / kNumRuns;
}

[[nodiscard]] std::expected<std::chrono::nanoseconds, BenchmarkError>
BenchmarkMatrixMultiplication(Backend backend) {