Files
mfem/tests/benchmarks/bench_assembly_levels.cpp
T

398 lines
13 KiB
C++

// 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
/*
This benchmark is inspired by the CEED's bake-off problems to benchmark the
performance of both the setup and the action of the different levels of
assembly in MFEM, namely: partial assembly (PA), element assembly (EA), and
full assembly (FA).
See: ceed.exascaleproject.org/bps and github.com/CEED/benchmarks
* --benchmark_filter=[SetupBP/BP/BK][1-6][PARTIAL/ELEMENT/FULL]/[1-max_order]
* --benchmark_context=device=[cpu/cuda/hip]
*/
// The maximum polynomial order used for benchmarking
const int max_order = 6;
// The maximum number of dofs for benchmarking
const int max_dofs = 1e7;
struct BakeOff
{
const AssemblyLevel assembly;
const int N, p, q, dim = 3;
Mesh mesh;
H1_FECollection fec;
FiniteElementSpace fes;
const Geometry::Type geom_type;
IntegrationRules IntRulesGLL;
const IntegrationRule *irGLL;
const IntegrationRule *ir;
ConstantCoefficient one;
const int dofs;
GridFunction x,y;
BilinearForm a;
double mdofs;
BakeOff(AssemblyLevel assembly, int p, int N, int vdim, bool GLL):
assembly(assembly),
N(N),
p(p),
q(2*p + (GLL?-1:3)),
mesh(Mesh::MakeCartesian3D(N,N,N,Element::HEXAHEDRON)),
fec(p, dim, BasisType::GaussLobatto),
fes(&mesh, &fec, vdim),
geom_type(mesh.GetTypicalElementGeometry()),
IntRulesGLL(0, Quadrature1D::GaussLobatto),
irGLL(&IntRulesGLL.Get(geom_type, q)),
ir(&IntRules.Get(geom_type, q)),
one(1.0),
dofs(fes.GetTrueVSize()),
x(&fes),
y(&fes),
a(&fes),
mdofs(0.0) {}
/// @brief Heuristic to evaluate if the case will run out of memory
bool is_runnable() const
{
const long long int gB = 1073741824/8;
const int mem_size = Device::IsEnabled()?16:256;
const long long int max_mem = mem_size * gB;
const int num_elems = fes.GetNE();
long long int mem = num_elems * pow(p+1, dim+1) * 8;
if (assembly == AssemblyLevel::ELEMENT)
{
mem += num_elems * pow(p+1, 2*dim) * 8;
}
if (assembly == AssemblyLevel::FULL)
{
mem += 3 * num_elems * pow(p+1, 2*dim) * 8;
}
// std::cout << "mem = " << mem << " , max_mem = " << max_mem << std::endl;
return mem < max_mem;
}
virtual void setup() = 0;
void benchmark_setup()
{
setup();
MFEM_DEVICE_SYNC;
mdofs += MDofs();
}
virtual void benchmark_action() = 0;
double SumMdofs() const { return mdofs; }
double MDofs() const { return 1e-6 * dofs; }
};
/// Bake-off Problems (BPs)
template<typename BFI, int VDIM = 1, bool GLL = false>
struct Problem: public BakeOff
{
const double rtol = 1e-12;
const int max_it = 32;
const int print_lvl = -1;
Array<int> ess_tdof_list;
Array<int> ess_bdr;
LinearForm b;
OperatorPtr A;
Vector B, X;
CGSolver cg;
Problem(AssemblyLevel assembly, int order, int N):
BakeOff(assembly,order,N,VDIM,GLL),
ess_bdr(mesh.bdr_attributes.Max()),
b(&fes)
{
if (is_runnable())
{
ess_bdr = 1;
fes.GetEssentialTrueDofs(ess_bdr,ess_tdof_list);
b.AddDomainIntegrator(new DomainLFIntegrator(one));
b.Assemble();
a.SetAssemblyLevel(assembly);
a.AddDomainIntegrator(new BFI(one, GLL?irGLL:ir));
a.Assemble();
a.Mult(x, y);
a.FormLinearSystem(ess_tdof_list, x, b, A, X, B);
cg.SetRelTol(rtol);
cg.SetOperator(*A);
cg.SetMaxIter(max_it);
cg.SetPrintLevel(print_lvl);
cg.iterative_mode = false;
MFEM_DEVICE_SYNC;
}
}
void setup() override
{
a.Assemble();
}
void benchmark_action() override
{
cg.Mult(B,X);
MFEM_DEVICE_SYNC;
mdofs += MDofs() * cg.GetNumIterations();
}
};
/// Bake-off inspired Problems (BPs) benchmarks for both the setup and action.
#define BakeOff_Problem(assembly,i,Kernel,VDIM,p_eq_q)\
static void SetupBP##i##assembly(bm::State &state){\
const int dim = 3;\
const int p = state.range(1);\
const int target_dofs = state.range(0);\
const int elem_dofs = pow(p+1, dim);\
const int N = pow(target_dofs / elem_dofs, 1.0/dim) + 1;\
Problem<Kernel##Integrator,VDIM,p_eq_q> ker(AssemblyLevel::assembly, p, N);\
if ( !ker.is_runnable() ) { state.SkipWithError("MAX_MEM"); }\
ker.setup();\
while (state.KeepRunning()) { ker.benchmark_setup(); }\
state.counters["MDof/s"] = bm::Counter(ker.SumMdofs(), bm::Counter::kIsRate);\
state.counters["Dofs"] = bm::Counter(ker.dofs, bm::Counter::kDefaults);\
state.counters["Order"] = bm::Counter(ker.p);\
state.counters["Assembly"] = (int)AssemblyLevel::assembly;}\
BENCHMARK(SetupBP##i##assembly)->ArgsProduct({\
benchmark::CreateRange(1024, max_dofs, /*step=*/2),\
benchmark::CreateDenseRange(1, max_order, /*step=*/1)\
})->Unit(bm::kMillisecond);\
static void BP##i##assembly(bm::State &state){\
const int dim = 3;\
const int p = state.range(1);\
const int target_dofs = state.range(0);\
const int elem_dofs = pow(p+1, dim);\
const int N = pow(target_dofs / elem_dofs, 1.0/dim) + 1;\
Problem<Kernel##Integrator,VDIM,p_eq_q> ker(AssemblyLevel::assembly, p, N);\
if ( !ker.is_runnable() ) { state.SkipWithError("MAX_MEM"); }\
while (state.KeepRunning()) { ker.benchmark_action(); }\
state.counters["MDof/s"] = bm::Counter(ker.SumMdofs(), bm::Counter::kIsRate);\
state.counters["Dofs"] = bm::Counter(ker.dofs, bm::Counter::kDefaults);\
state.counters["Order"] = bm::Counter(ker.p);\
state.counters["Assembly"] = (int)AssemblyLevel::assembly;}\
BENCHMARK(BP##i##assembly)->ArgsProduct({\
benchmark::CreateRange(1024, max_dofs, /*step=*/2),\
benchmark::CreateDenseRange(1, max_order, /*step=*/1)\
})->Unit(bm::kMillisecond);
// PARTIAL:
/// BP1: scalar PCG with mass matrix, q=p+2
BakeOff_Problem(PARTIAL,1,Mass,1,false)
/// BP2: vector PCG with mass matrix, q=p+2
BakeOff_Problem(PARTIAL,2,VectorMass,3,false)
/// BP3: scalar PCG with stiffness matrix, q=p+2
BakeOff_Problem(PARTIAL,3,Diffusion,1,false)
/// BP4: vector PCG with stiffness matrix, q=p+2
BakeOff_Problem(PARTIAL,4,VectorDiffusion,3,false)
/// BP5: scalar PCG with stiffness matrix, q=p+1
BakeOff_Problem(PARTIAL,5,Diffusion,1,true)
/// BP6: vector PCG with stiffness matrix, q=p+1
BakeOff_Problem(PARTIAL,6,VectorDiffusion,3,true)
// ELEMENT:
/// BP1: scalar PCG with mass matrix, q=p+2
BakeOff_Problem(ELEMENT,1,Mass,1,false)
/// BP2: vector PCG with mass matrix, q=p+2
// BakeOff_Problem(ELEMENT,2,VectorMass,3,false) // Not yet implemented
/// BP3: scalar PCG with stiffness matrix, q=p+2
BakeOff_Problem(ELEMENT,3,Diffusion,1,false)
/// BP4: vector PCG with stiffness matrix, q=p+2
// BakeOff_Problem(ELEMENT,4,VectorDiffusion,3,false) // Not yet implemented
/// BP5: scalar PCG with stiffness matrix, q=p+1
BakeOff_Problem(ELEMENT,5,Diffusion,1,true)
/// BP6: vector PCG with stiffness matrix, q=p+1
// BakeOff_Problem(ELEMENT,6,VectorDiffusion,3,true) // Not yet implemented
// FULL:
/// BP1: scalar PCG with mass matrix, q=p+2
BakeOff_Problem(FULL,1,Mass,1,false)
/// BP2: vector PCG with mass matrix, q=p+2
// BakeOff_Problem(FULL,2,VectorMass,3,false) // Not yet implemented
/// BP3: scalar PCG with stiffness matrix, q=p+2
BakeOff_Problem(FULL,3,Diffusion,1,false)
/// BP4: vector PCG with stiffness matrix, q=p+2
// BakeOff_Problem(FULL,4,VectorDiffusion,3,false) // Not yet implemented
/// BP5: scalar PCG with stiffness matrix, q=p+1
BakeOff_Problem(FULL,5,Diffusion,1,true)
/// BP6: vector PCG with stiffness matrix, q=p+1
// BakeOff_Problem(FULL,6,VectorDiffusion,3,true) // Not yet implemented
/// Bake-off Kernels (BKs)
template <typename BFI, int VDIM = 1, bool GLL = false>
struct Kernel: public BakeOff
{
GridFunction y;
Kernel(AssemblyLevel assembly, int order, int N)
: BakeOff(assembly,order,N,VDIM,GLL), y(&fes)
{
if (is_runnable())
{
x.Randomize(1);
a.SetAssemblyLevel(assembly);
a.AddDomainIntegrator(new BFI(one, GLL?irGLL:ir));
a.Assemble();
a.Mult(x, y);
MFEM_DEVICE_SYNC;
}
}
void setup() override
{
a.Assemble();
}
void benchmark_action() override
{
a.Mult(x, y);
MFEM_DEVICE_SYNC;
mdofs += MDofs();
}
};
/// CEED BKi inspired benchmark for the action
#define BakeOff_Kernel(assembly,i,KER,VDIM,GLL)\
static void BK##i##assembly(bm::State &state){\
const int dim = 3;\
const int p = state.range(1);\
const int target_dofs = state.range(0);\
const int elem_dofs = pow(p+1, dim);\
const int N = pow(target_dofs / elem_dofs, 1.0/dim) + 1;\
Kernel<KER##Integrator,VDIM,GLL> ker(AssemblyLevel::assembly, p, N);\
if ( !ker.is_runnable() ) { state.SkipWithError("MAX_MEM"); }\
while (state.KeepRunning()) { ker.benchmark_action(); }\
state.counters["MDof/s"] = bm::Counter(ker.SumMdofs(), bm::Counter::kIsRate);\
state.counters["Dofs"] = bm::Counter(ker.dofs, bm::Counter::kDefaults);\
state.counters["Order"] = bm::Counter(ker.p);\
state.counters["Assembly"] = (int)AssemblyLevel::assembly;}\
BENCHMARK(BK##i##assembly)->ArgsProduct({\
benchmark::CreateRange(1024, max_dofs, /*step=*/2),\
benchmark::CreateDenseRange(1, max_order, /*step=*/1)\
})->Unit(bm::kMillisecond);\
// PARTIAL:
/// BK1PARTIAL: scalar E-vector-to-E-vector evaluation of mass matrix, q=p+2
BakeOff_Kernel(PARTIAL,1,Mass,1,false)
/// BK2PARTIAL: vector E-vector-to-E-vector evaluation of mass matrix, q=p+2
BakeOff_Kernel(PARTIAL,2,VectorMass,3,false)
/// BK3PARTIAL: scalar E-vector-to-E-vector evaluation of stiffness matrix, q=p+2
BakeOff_Kernel(PARTIAL,3,Diffusion,1,false)
/// BK4PARTIAL: vector E-vector-to-E-vector evaluation of stiffness matrix, q=p+2
BakeOff_Kernel(PARTIAL,4,VectorDiffusion,3,false)
/// BK5PARTIAL: scalar E-vector-to-E-vector evaluation of stiffness matrix, q=p+1
BakeOff_Kernel(PARTIAL,5,Diffusion,1,true)
/// BK6PARTIAL: vector E-vector-to-E-vector evaluation of stiffness matrix, q=p+1
BakeOff_Kernel(PARTIAL,6,VectorDiffusion,3,true)
// ELEMENT
/// BK1ELEMENT: scalar E-vector-to-E-vector evaluation of mass matrix, q=p+2
BakeOff_Kernel(ELEMENT,1,Mass,1,false)
/// BK2ELEMENT: vector E-vector-to-E-vector evaluation of mass matrix, q=p+2
// BakeOff_Kernel(ELEMENT,2,VectorMass,3,false) // Not yet implemented
/// BK3ELEMENT: scalar E-vector-to-E-vector evaluation of stiffness matrix, q=p+2
BakeOff_Kernel(ELEMENT,3,Diffusion,1,false)
/// BK4ELEMENT: vector E-vector-to-E-vector evaluation of stiffness matrix, q=p+2
// BakeOff_Kernel(ELEMENT,4,VectorDiffusion,3,false) // Not yet implemented
/// BK5ELEMENT: scalar E-vector-to-E-vector evaluation of stiffness matrix, q=p+1
BakeOff_Kernel(ELEMENT,5,Diffusion,1,true)
/// BK6ELEMENT: vector E-vector-to-E-vector evaluation of stiffness matrix, q=p+1
// BakeOff_Kernel(ELEMENT,6,VectorDiffusion,3,true) // Not yet implemented
// FULL
/// BK1FULL: scalar E-vector-to-E-vector evaluation of mass matrix, q=p+2
BakeOff_Kernel(FULL,1,Mass,1,false)
/// BK2FULL: vector E-vector-to-E-vector evaluation of mass matrix, q=p+2
// BakeOff_Kernel(FULL,2,VectorMass,3,false) // Not yet implemented
/// BK3FULL: scalar E-vector-to-E-vector evaluation of stiffness matrix, q=p+2
BakeOff_Kernel(FULL,3,Diffusion,1,false)
/// BK4FULL: vector E-vector-to-E-vector evaluation of stiffness matrix, q=p+2
// BakeOff_Kernel(FULL,4,VectorDiffusion,3,false) // Not yet implemented
/// BK5FULL: scalar E-vector-to-E-vector evaluation of stiffness matrix, q=p+1
BakeOff_Kernel(FULL,5,Diffusion,1,true)
/// BK6FULL: vector E-vector-to-E-vector evaluation of stiffness matrix, q=p+1
// BakeOff_Kernel(FULL,6,VectorDiffusion,3,true) // Not yet implemented
/**
* @brief main entry point
* --benchmark_filter=BK1/6
* --benchmark_context=device=cpu
*/
int main(int argc, char *argv[])
{
bm::ConsoleReporter CR;
bm::Initialize(&argc, argv);
// Device setup, cpu by default
std::string device_config = "cpu";
auto global_context = bmi::GetGlobalContext();
if (global_context != nullptr)
{
const auto device = global_context->find("device");
if (device != global_context->end())
{
mfem::out << device->first << " : " << device->second << std::endl;
device_config = device->second;
}
}
Device device(device_config.c_str());
device.Print();
if (bm::ReportUnrecognizedArguments(argc, argv)) { return 1; }
bm::RunSpecifiedBenchmarks(&CR);
return 0;
}
#endif // MFEM_USE_BENCHMARK