Skip to content

Alltoall embedded timings #42

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
shannong wants to merge 8 commits into
base: locality_alltoalls2
Choose a base branch
from
Open

Alltoall embedded timings #42

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
8acaa9a
adding timing and estimation functions
shannong Aug 4, 2025
ad91411
adding internal timings to new algorithms
shannong Aug 4, 2025
482bc3c
fixing compile issues
shannong Aug 4, 2025
5c45770
fixed function calls
shannong Aug 7, 2025
c600f9e
fixing internal timings - deadlocks, improving print statements
shannong Aug 7, 2025
fc2dac5
resetting max size
shannong Aug 7, 2025
ba8a733
fixing seg fault in internal timings due to using the wrong buffer fo…
shannong Aug 8, 2025
09c0572
cleaning up output
shannong Aug 11, 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
82 changes: 43 additions & 39 deletions benchmarks/alltoall_topology.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,65 +61,69 @@ void print_alltoalls(int max_p, const T* sendbuf,
double time;

using F = int (*)(const void*, int, MPI_Datatype, void*, int, MPI_Datatype, _MPIX_Comm*);
std::vector<F> alltoall_funcs = {alltoall_pairwise, alltoall_nonblocking, alltoall_hierarchical, alltoall_node_aware, alltoall_hierarchical_nb, alltoall_node_aware_nb};
std::vector<const char*> names = {"Pairwise", "NonBlocking", "Pairwise Hierarchical", "Pairwise Node Aware", "Nonblocking Hierarchical", "Nonblocking Node Aware"};
std::vector<F> alltoall_funcs = {alltoall_hierarchical, alltoall_node_aware, alltoall_hierarchical_nb, alltoall_node_aware_nb};
std::vector<const char*> names = {"Pairwise Hierarchical", "Pairwise Node Aware", "Nonblocking Hierarchical", "Nonblocking Node Aware"};

std::vector<F> multileader_funcs = { alltoall_multileader, alltoall_locality_aware, alltoall_multileader_locality, alltoall_multileader_nb, alltoall_locality_aware_nb, alltoall_multileader_locality_nb};
std::vector<const char*> multileader_names = {"Pairwise Multileader", "Pairwise Locality Aware", "Pairwise Multileader Locality", "Nonblocking Multileader", "Nonblocking Locality Aware", "Nonblocking Multileader Locality"};

for (int i = 0; i < max_p; i++)
for (int i = 0; i < max_p; i++)
{
int s = pow(2, i);
int s = pow(2, i);

if (rank == 0) printf("Size %d\n", s);
if (rank == 0) printf("Size %d\n", s);

// Standard PMPI Alltoall (system MPI)

PMPI_Alltoall(sendbuf, s, sendtype, recvbuf, s, recvtype, comm->global_comm);
std::memcpy(recvbuf_std, recvbuf, s*sizeof(T));
time = test_alltoall(PMPI_Alltoall, sendbuf, s, sendtype,
recvbuf, s, recvtype, comm->global_comm);
recvbuf, s, recvtype, comm->global_comm);
if (rank == 0) printf("PMPI: %e\n", time);

// MPI Advance Alltoall Functions (not multileader)
for (int idx = 0; idx < alltoall_funcs.size(); idx++)
{
alltoall_funcs[idx](sendbuf, s, sendtype, recvbuf, s, recvtype, comm);
for (int j = 0; j < s; j++)
if (fabs(recvbuf_std[j] - recvbuf[j]) > 1e-06)
{
printf("DIFF RESULTS %d vs %d\n", recvbuf_std[j], recvbuf[j]);
MPI_Abort(comm->global_comm, -1);
}
time = test_alltoall(alltoall_funcs[idx], sendbuf, s, sendtype,
recvbuf, s, recvtype, comm);
if (rank == 0) printf("%s: %e\n", names[idx], time);
if (rank == 0)
printf("Testing %s\n", names[idx]);
alltoall_funcs[idx](sendbuf, s, sendtype, recvbuf, s, recvtype, comm);
//for (int j = 0; j < s; j++)
// if (fabs(recvbuf_std[j] - recvbuf[j]) > 1e-06)
// {
// printf("DIFF RESULTS %d vs %d\n", recvbuf_std[j], recvbuf[j]);
// MPI_Abort(comm->global_comm, -1);
// }
time = test_alltoall(alltoall_funcs[idx], sendbuf, s, sendtype,
recvbuf, s, recvtype, comm);
if (rank == 0) printf("%s: %e\n", names[idx], time);
}

// MPI Advance Multileader Alltoall Functions
if (rank == 0)
printf("Testing multileader functions\n");
std::vector<int> procs_per_leader_list = {4, 8, 16};
for (int ctr = 0; ctr < procs_per_leader_list.size(); ctr++)
{
int n_procs = procs_per_leader_list[ctr];
if (ppn < n_procs)
break;
MPIX_Comm_leader_init(comm, n_procs);

for (int idx = 0; idx < multileader_funcs.size(); idx++)
{
multileader_funcs[idx](sendbuf, s, sendtype, recvbuf, s, recvtype, comm);
for (int j = 0; j < s; j++)
if (fabs(recvbuf_std[j] - recvbuf[j]) > 1e-06)
{
printf("DIFF RESULTS %d vs %d\n", recvbuf_std[j], recvbuf[j]);
MPI_Abort(comm->global_comm, -1);
}
time = test_alltoall(multileader_funcs[idx], sendbuf, s, sendtype,
recvbuf, s, recvtype, comm);
if (rank == 0) printf("%s, %d procs per leader: %e\n", multileader_names[idx], n_procs, time);
}

MPIX_Comm_leader_free(comm);
}
int n_procs = procs_per_leader_list[ctr];
if (ppn < n_procs)
break;
MPIX_Comm_leader_init(comm, n_procs);

for (int idx = 0; idx < multileader_funcs.size(); idx++)
{
// for (int j = 0; j < s; j++)
//if (fabs(recvbuf_std[j] - recvbuf[j]) > 1e-06)
//{
// printf("DIFF RESULTS %d vs %d\n", recvbuf_std[j], recvbuf[j]);
// MPI_Abort(comm->global_comm, -1);
// }
if (rank == 0) printf("%s, %d procs per leader: %e\n", multileader_names[idx], n_procs, time);
time = test_alltoall(multileader_funcs[idx], sendbuf, s, sendtype, recvbuf, s, recvtype, comm);

}

MPIX_Comm_leader_free(comm);
}
}

}
Expand All @@ -129,7 +133,7 @@ int main(int argc, char* argv[])
{
MPI_Init(&argc, &argv);

int max_p = 15;
int max_p = 11;
int max_size = pow(2, max_p);

MPIX_Comm* xcomm;
Expand All @@ -138,7 +142,7 @@ int main(int argc, char* argv[])
int rank, num_procs;
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &num_procs);

MPIX_Comm_topo_init(xcomm);

int local_rank, ppn;
Expand Down
Loading