mfem/tests/benchmarks/bench_vector.cpp

// Copyright (c) 2010-2025, Lawrence Livermore National Security, LLC. Produced
// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
// LICENSE and NOTICE for details. LLNL-CODE-806117.
//
// This file is part of the MFEM library. For more information and source code
// availability visit https://mfem.org.
//
// MFEM is free software; you can redistribute it and/or modify it under the
// terms of the BSD-3 license. We welcome feedback and contributions, see file
// CONTRIBUTING.md for details.

#include "bench.hpp"

#ifdef MFEM_USE_BENCHMARK

using namespace mfem;

// Default macro to register the tests
#define MFEM_VECTOR_BENCHMARK(x) BENCHMARK(x)->RangeMultiplier(4)->Range(1,KB);

// benchmark::CLASS generator with specific prefix
namespace benchmark
{

#define GENERATE_CLASS(CLASS, PREFIX)\
class CLASS{\
protected:\
Memory<double> data; int size;\
public:\
explicit CLASS(int s);\
PREFIX ~CLASS();\
PREFIX void UseDevice(bool use_dev) const { data.UseDevice(use_dev); }\
PREFIX bool UseDevice() const { return data.UseDevice(); }\
PREFIX const double *Read(bool on_dev = true) const\
{ return mfem::Read(data, size, on_dev); }\
PREFIX double *Write(bool on_dev = true)\
{ return mfem::Write(data, size, on_dev); }\
PREFIX double *ReadWrite(bool on_dev = true)\
{ return mfem::ReadWrite(data, size, on_dev); }\
CLASS &operator=(double value);\
CLASS &operator+=(const CLASS &v);\
};\
inline CLASS::CLASS(int s){\
   if (s > 0) { size = s; data.New(s); }\
   else { size = 0; data.Reset(); }\
}\
inline CLASS::~CLASS() { data.Delete(); }\
inline CLASS &CLASS::operator=(double value){\
   const bool use_dev = UseDevice()/*true*/;\
   const int N = size;\
   auto y = /*mfem::Write(data, size, use_dev)*/Write(use_dev);\
   forall_switch(use_dev, N, [=] MFEM_HOST_DEVICE (int i) {y[i] = value;});\
   return *this;\
}\
inline CLASS &CLASS::operator+=(const CLASS &v){\
   MFEM_ASSERT(size == v.size, "incompatible Vectors!");\
   const bool use_dev = UseDevice() || v.UseDevice();\
   const int N = size;\
   auto y = ReadWrite(use_dev);\
   auto x = v.Read(use_dev);\
   forall_switch(use_dev, N, [=] MFEM_HOST_DEVICE (int i) {y[i] += x[i];});\
   return *this;\
}

GENERATE_CLASS(Vector,);
GENERATE_CLASS(Vector_Virtuals, virtual);
GENERATE_CLASS(Vector_Virtuals_Inlined, virtual inline);

} // namespace benchmark


// Operator =

static void Vector_EQ_MFEM(benchmark::State& state)
{
   const size_t size = state.range(0);
   mfem::Vector x(size);
   for (auto _ : state) { x = M_PI; }
}
MFEM_VECTOR_BENCHMARK(Vector_EQ_MFEM);

static void Vector_EQ(benchmark::State& state)
{
   const size_t size = state.range(0);
   benchmark::Vector x(size);
   for (auto _ : state) { x = M_PI; }
}
MFEM_VECTOR_BENCHMARK(Vector_EQ);

static void Vector_EQ_Virtuals(benchmark::State& state)
{
   const size_t size = state.range(0);
   benchmark::Vector_Virtuals x(size);
   for (auto _ : state) { x = M_PI; }
}
MFEM_VECTOR_BENCHMARK(Vector_EQ_Virtuals);

static void Vector_EQ_Virtuals_Inlined(benchmark::State& state)
{
   const size_t size = state.range(0);
   benchmark::Vector_Virtuals_Inlined x(size);
   for (auto _ : state) { x = M_PI; }
}
MFEM_VECTOR_BENCHMARK(Vector_EQ_Virtuals_Inlined);


// Operator +=

static void Vector_PE_MFEM(benchmark::State& state)
{
   const size_t size = state.range(0);
   mfem::Vector x(size), y(size);
   x = M_PI; y = M_E;
   for (auto _ : state) { x += y; }
}
MFEM_VECTOR_BENCHMARK(Vector_PE_MFEM);

static void Vector_PE_MFEM_64(benchmark::State& state)
{
   const size_t size = state.range(0);
   mfem::Vector x(size, MemoryType::HOST_64);
   mfem::Vector y(size, MemoryType::HOST_64);
   x = M_PI; y = M_E;
   for (auto _ : state) { x += y; }
}
MFEM_VECTOR_BENCHMARK(Vector_PE_MFEM_64);

static void Vector_PE(benchmark::State& state)
{
   const size_t size = state.range(0);
   benchmark::Vector x(size); x = M_PI;
   benchmark::Vector y(size); y = M_E;
   for (auto _ : state) { x += y; }
}
MFEM_VECTOR_BENCHMARK(Vector_PE);

static void Vector_PE_Virtuals(benchmark::State& state)
{
   const size_t size = state.range(0);
   benchmark::Vector_Virtuals x(size); x = M_PI;
   benchmark::Vector_Virtuals y(size); y = M_E;
   for (auto _ : state) { x += y; }
}
MFEM_VECTOR_BENCHMARK(Vector_PE_Virtuals);

static void Vector_Virtuals_Inlined_PE(benchmark::State& state)
{
   const size_t size = state.range(0);
   benchmark::Vector_Virtuals_Inlined x(size); x = M_PI;
   benchmark::Vector_Virtuals_Inlined y(size); y = M_E;
   for (auto _ : state) { x += y; }
}
MFEM_VECTOR_BENCHMARK(Vector_Virtuals_Inlined_PE);

// Base class
struct Base
{
   void nop()
   {
      int unused;
      benchmark::DoNotOptimize(unused);
      asm volatile("nop" : "=r" (unused));
   }
};

static void Base(benchmark::State& state)
{
   struct Base b;
   for (auto _ : state) { b.nop(); }
}
MFEM_VECTOR_BENCHMARK(Base);

class Base_Virtuals
{
public:
   virtual void nop()
   {
      int unused;
      benchmark::DoNotOptimize(unused);
      asm volatile("nop" : "=r" (unused));
   }
};

static void Base_Virtuals_inlined(benchmark::State& state)
{
   // it will be resolved at compile time, so it can be inlined
   Base_Virtuals b;
   for (auto _ : state) { b.nop(); }
}
MFEM_VECTOR_BENCHMARK(Base_Virtuals_inlined);

// Base dummy BMDerivedVector class
class Base_Virtuals_Derived: public Base_Virtuals
{
public:
   void nop()
   {
      int unused;
      benchmark::DoNotOptimize(unused);
      asm volatile("nop" : "=r" (unused));
   }
};

static void Base_Virtuals_Derived_not_inlined(benchmark::State& state)
{
   // cannot be inlined through the pointer
   Base_Virtuals *ptr = new Base_Virtuals_Derived();
   for (auto _ : state) { ptr->nop(); }
}
MFEM_VECTOR_BENCHMARK(Base_Virtuals_Derived_not_inlined);

// --benchmark_filter=all
// --benchmark_filter=Vector_PE_MFEM
int main(int argc, char *argv[])
{
   mfem::Reporter mfem_reporter;
   ::benchmark::Initialize(&argc, argv);
   if (::benchmark::ReportUnrecognizedArguments(argc, argv)) { return 1; }
   ::benchmark::RunSpecifiedBenchmarks(&mfem_reporter);
   return 0;
}

#endif // MFEM_USE_BENCHMARK