Skip to content

[bench] add dense (add|sub)_diag to blas #1909

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

Open
wants to merge 1 commit into
base: develop
Choose a base branch
from
Open

[bench] add dense (add|sub)_diag to blas #1909

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
10 changes: 10 additions & 0 deletions benchmark/blas/blas.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -77,6 +77,16 @@ std::map<std::string, std::function<std::unique_ptr<BenchmarkOperation>(
exec, Generator{}, dims.n, dims.k, dims.m, dims.stride_A,
dims.stride_B, dims.stride_C);
}},
{"add_diag",
[](std::shared_ptr<const gko::Executor> exec, dimensions dims) {
return std::make_unique<AddDiagOperation<Generator>>(
exec, Generator{}, dims.n, dims.stride_A);
}},
{"sub_diag",
[](std::shared_ptr<const gko::Executor> exec, dimensions dims) {
return std::make_unique<SubDiagOperation<Generator>>(
exec, Generator{}, dims.n, dims.stride_A);
}},
{"prefix_sum32",
[](std::shared_ptr<const gko::Executor> exec, dimensions dims) {
return std::make_unique<PrefixSumOperation<gko::int32>>(exec,
Expand Down
74 changes: 74 additions & 0 deletions benchmark/blas/blas_common.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,6 +36,8 @@ DEFINE_string(
" norm (a = sqrt(x' * x)),\n"
" mm (C = A * B),\n"
" gemm (C = a * A * B + b * C)\n"
" add_diag (A = A + a * D)\n"
" sub_diag (A = A - a * D)\n"
"Non-numerical algorithms:\n"
" prefix_sum32 (x_i <- sum_{j=0}^{i-1} x_i, 32 bit indices)\n"
" prefix_sum64 ( 64 bit indices)\n"
Expand Down Expand Up @@ -373,6 +375,78 @@ class AdvancedApplyOperation : public BenchmarkOperation {
};


template <typename Generator>
class AddDiagOperation : public BenchmarkOperation {
public:
AddDiagOperation(std::shared_ptr<const gko::Executor> exec,
const Generator& generator, gko::size_type n,
gko::size_type stride)
{
// Since dense distributed matrices are not supported we can use
// local_size == global_size
A_ = generator.create_multi_vector_strided(exec, gko::dim<2>{n, n},
gko::dim<2>{n, n}, stride);
D_ = gko::matrix::Diagonal<etype>::create(exec, n);
alpha_ = gko::matrix::Dense<etype>::create(exec, gko::dim<2>{1, 1});

as_vector<Generator>(A_)->fill(1);
D_->read(gko::matrix_data<etype, itype>::diag(gko::dim<2>{n, n},
etype{2.2}));
alpha_->fill(1);
}

gko::size_type get_flops() const override { return A_->get_size()[0] * 2; }

gko::size_type get_memory() const override
{
return A_->get_size()[0] * 3 * sizeof(etype);
}

void run() override { as_vector<Generator>(A_)->add_scaled(alpha_, D_); }

private:
std::unique_ptr<gko::matrix::Dense<etype>> alpha_;
std::unique_ptr<gko::LinOp> A_;
std::unique_ptr<gko::matrix::Diagonal<etype>> D_;
Copy link
Member

Choose a reason for hiding this comment

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

we need to make it to snake_case according to https://github.com/ginkgo-project/ginkgo/wiki/Contributing-guidelines#naming-scheme

Copy link
Member Author

Choose a reason for hiding this comment

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

I think the naming scheme here should match the operation description in the CLI flag. I adjusted it accordingly.

};


template <typename Generator>
class SubDiagOperation : public BenchmarkOperation {
public:
SubDiagOperation(std::shared_ptr<const gko::Executor> exec,
const Generator& generator, gko::size_type n,
gko::size_type stride)
{
// Since dense distributed matrices are not supported we can use
// local_size == global_size
A_ = generator.create_multi_vector_strided(exec, gko::dim<2>{n, n},
gko::dim<2>{n, n}, stride);
D_ = gko::matrix::Diagonal<etype>::create(exec, n);
alpha_ = gko::matrix::Dense<etype>::create(exec, gko::dim<2>{1, 1});

as_vector<Generator>(A_)->fill(1);
D_->read(gko::matrix_data<etype, itype>::diag(gko::dim<2>{n, n},
etype{2.2}));
alpha_->fill(1);
}

gko::size_type get_flops() const override { return A_->get_size()[0] * 2; }

gko::size_type get_memory() const override
{
return A_->get_size()[0] * 3 * sizeof(etype);
}

void run() override { as_vector<Generator>(A_)->sub_scaled(alpha_, D_); }

private:
std::unique_ptr<gko::matrix::Dense<etype>> alpha_;
std::unique_ptr<gko::LinOp> A_;
std::unique_ptr<gko::matrix::Diagonal<etype>> D_;
};


GKO_REGISTER_OPERATION(prefix_sum_nonnegative,
components::prefix_sum_nonnegative);

Expand Down
Loading