Skip to content

add benchmark_from_scratch flag #1881

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Aug 15, 2025
Merged

add benchmark_from_scratch flag #1881

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
c9a84ff
add benchmark_from_scratch flag
yhmtsai Jul 7, 2025
972e549
fix typo and no warmup iter issue
yhmtsai Jul 30, 2025
4fa322c
ensure the logger log proper stuff
yhmtsai Jul 30, 2025
ea5cba2
fix the benchmark not from scratch still use warm up as iter limit
yhmtsai Aug 4, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 4 additions & 2 deletions benchmark/preconditioner/preconditioner.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -225,7 +225,6 @@ struct PreconditionerBenchmark : Benchmark<preconditioner_benchmark_state> {
auto x_clone = clone(state.x);
auto precond = precond_factory.at(decoded_precond_name)(exec);

std::unique_ptr<gko::LinOp> precond_op;
{
auto gen_logger = create_operations_logger(
FLAGS_gpu_timer, FLAGS_nested_names, exec,
Expand All @@ -236,14 +235,17 @@ struct PreconditionerBenchmark : Benchmark<preconditioner_benchmark_state> {
exec->get_master()->add_logger(gen_logger);
}
for (auto i = 0u; i < ic_gen.get_num_repetitions(); ++i) {
precond_op = precond->generate(state.system_matrix);
auto precond_op = precond->generate(state.system_matrix);
}
if (exec->get_master() != exec) {
exec->get_master()->remove_logger(gen_logger);
}
exec->remove_logger(gen_logger);
}

// generate it for apply usage
auto precond_op = precond->generate(state.system_matrix);

auto apply_logger = create_operations_logger(
FLAGS_gpu_timer, FLAGS_nested_names, exec,
precond_case["apply"]["components"],
Expand Down
62 changes: 45 additions & 17 deletions benchmark/solver/solver_common.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,11 @@ DEFINE_bool(
rel_residual, false,
"Use relative residual instead of residual reduction stopping criterion");

DEFINE_bool(benchmark_from_scratch, false,
"benchmark the solver from scratch everytime which requires "
"workspace initialization everytime. When this is true, the "
"repetition progress will use the solver generated additionally.");

DEFINE_string(solvers, "cg",
"A comma-separated list of solvers to run. "
"Supported values are: bicgstab, bicg, cb_gmres_keep, "
Expand Down Expand Up @@ -499,7 +504,6 @@ struct SolverBenchmark : Benchmark<solver_benchmark_state<Generator>> {
if (FLAGS_detailed && !FLAGS_overhead) {
// slow run, get the time of each functions
auto x_clone = clone(state.x);

{
auto gen_logger = create_operations_logger(
FLAGS_gpu_timer, FLAGS_nested_names, exec,
Expand All @@ -509,19 +513,27 @@ struct SolverBenchmark : Benchmark<solver_benchmark_state<Generator>> {
exec->get_master()->add_logger(gen_logger);
}

auto precond = precond_factory.at(precond_name)(exec);
solver = generate_solver(exec, give(precond), solver_name,
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

free entry in the previous output is deleting the workspace from previous warmup apply not the generation. It introduces the changes for test output

FLAGS_max_iters)
->generate(state.system_matrix);
{
auto precond = precond_factory.at(precond_name)(exec);
auto solver = generate_solver(exec, give(precond),
solver_name, FLAGS_max_iters)
->generate(state.system_matrix);
}

exec->remove_logger(gen_logger);
if (exec != exec->get_master()) {
exec->get_master()->remove_logger(gen_logger);
}
}

if (auto prec =
dynamic_cast<const gko::Preconditionable*>(solver.get())) {
// generate it for apply usage
auto precond = precond_factory.at(precond_name)(exec);
auto detailed_solver = generate_solver(exec, give(precond),
solver_name, FLAGS_max_iters)
->generate(state.system_matrix);

if (auto prec = dynamic_cast<const gko::Preconditionable*>(
detailed_solver.get())) {
solver_case["preconditioner"] = json::object();
write_precond_info(
clone(exec->get_master(), prec->get_preconditioner()).get(),
Expand All @@ -537,7 +549,7 @@ struct SolverBenchmark : Benchmark<solver_benchmark_state<Generator>> {
exec->get_master()->add_logger(apply_logger);
}

solver->apply(state.b, x_clone);
detailed_solver->apply(state.b, x_clone);

exec->remove_logger(apply_logger);
if (exec != exec->get_master()) {
Expand All @@ -554,8 +566,8 @@ struct SolverBenchmark : Benchmark<solver_benchmark_state<Generator>> {
solver_case["true_residuals"],
solver_case["implicit_residuals"],
solver_case["iteration_timestamps"]);
solver->add_logger(res_logger);
solver->apply(state.b, x_clone);
detailed_solver->add_logger(res_logger);
detailed_solver->apply(state.b, x_clone);
if (!res_logger->has_implicit_res_norms()) {
solver_case.erase("implicit_residuals");
}
Expand All @@ -568,17 +580,33 @@ struct SolverBenchmark : Benchmark<solver_benchmark_state<Generator>> {
auto generate_timer = get_timer(exec, FLAGS_gpu_timer);
auto apply_timer = ic.get_timer();
auto x_clone = clone(state.x);
// if we benchmark from scratch, we generate it here and do operations
// once. we can not rely on the warmup one because it use different
// iteration criterion.
if (FLAGS_benchmark_from_scratch) {
auto precond = precond_factory.at(precond_name)(exec);
solver = gko::share(generate_solver(exec, give(precond),
solver_name, FLAGS_max_iters)
->generate(state.system_matrix));
solver->apply(state.b, x_clone);
}
for (auto status : ic.run(false)) {
auto range = annotate("repetition");
x_clone = clone(state.x);

exec->synchronize();
generate_timer->tic();
auto precond = precond_factory.at(precond_name)(exec);
solver = generate_solver(exec, give(precond), solver_name,
FLAGS_max_iters)
->generate(state.system_matrix);
generate_timer->toc();
{
exec->synchronize();
generate_timer->tic();
auto precond = precond_factory.at(precond_name)(exec);
auto generated_solver =
gko::share(generate_solver(exec, give(precond), solver_name,
FLAGS_max_iters)
->generate(state.system_matrix));
generate_timer->toc();
if (FLAGS_benchmark_from_scratch || !solver) {
solver = generated_solver;
}
}

exec->synchronize();
if (ic.get_num_repetitions() == 0) {
Expand Down
1 change: 0 additions & 1 deletion benchmark/test/reference/distributed_solver.matrix.stdout
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,6 @@
"generate": {
"components": {
"generate(<typename>)": 1.0,
"free": 1.0,
"overhead": 1.0
},
"time": 1.0
Expand Down
1 change: 0 additions & 1 deletion benchmark/test/reference/distributed_solver.simple.stdout
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,6 @@
"generate": {
"components": {
"generate(<typename>)": 1.0,
"free": 1.0,
"overhead": 1.0
},
"time": 1.0
Expand Down
1 change: 0 additions & 1 deletion benchmark/test/reference/distributed_solver_dcomplex.simple.stdout
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,6 @@
"generate": {
"components": {
"generate(<typename>)": 1.0,
"free": 1.0,
"overhead": 1.0
},
"time": 1.0
Expand Down
1 change: 0 additions & 1 deletion benchmark/test/reference/solver.matrix.stdout
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,6 @@
"generate": {
"components": {
"generate(<typename>)": 1.0,
"free": 1.0,
"overhead": 1.0
},
"time": 1.0
Expand Down
1 change: 0 additions & 1 deletion benchmark/test/reference/solver.reordered.stdout
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,6 @@
"generate": {
"components": {
"generate(<typename>)": 1.0,
"free": 1.0,
"overhead": 1.0
},
"time": 1.0
Expand Down
1 change: 0 additions & 1 deletion benchmark/test/reference/solver.simple.stdout
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,6 @@
"generate": {
"components": {
"generate(<typename>)": 1.0,
"free": 1.0,
"overhead": 1.0
},
"time": 1.0
Expand Down
1 change: 0 additions & 1 deletion benchmark/test/reference/solver_dcomplex.simple.stdout
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,6 @@
"generate": {
"components": {
"generate(<typename>)": 1.0,
"free": 1.0,
"overhead": 1.0
},
"time": 1.0
Expand Down
Loading