Skip to content
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

Enable concurrent broadcasts in update_edge_partition_minor_property() #2413

Merged
merged 6 commits into from
Jul 21, 2022
Merged

Enable concurrent broadcasts in update_edge_partition_minor_property() #2413

Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
34f3643
enable concurrent broadcasts in update_edge_partition_src_dst_prpoerty
seunghwak Jul 14, 2022
659f049
added device_group_start|end
seunghwak Jul 15, 2022
fb3824d
clang-format
seunghwak Jul 19, 2022
93896fd
inline device_group_start/end
seunghwak Jul 19, 2022
80fff08
fix compiler warning
seunghwak Jul 19, 2022
60eb174
Merge branch 'branch-22.08' of github.com:rapidsai/cugraph into enh_p…
seunghwak Jul 19, 2022
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
40 changes: 3 additions & 37 deletions cpp/src/prims/update_edge_partition_src_dst_property.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -294,8 +294,6 @@ void update_edge_partition_minor_property(
key_offsets = *(graph_view.local_sorted_unique_edge_dst_vertex_partition_offsets());
}

#if 0 // FIXME: should not directly call ncclGroupStart(), ncclGroupEnd(), enable after adding a
// raft::comms mechanism to group broadcast operations
// memory footprint vs parallelism trade-off
// memory requirement per loop is
// (V/comm_size) * sizeof(value_t)
Expand Down Expand Up @@ -327,10 +325,7 @@ void update_edge_partition_minor_property(
}

for (size_t round = 0; round < num_rounds; ++round) {
// FIXME: better use a higher-level interface than directly invoking NCCL functions.
if (ncclGroupStart() != ncclSuccess) {
CUGRAPH_FAIL("ncclGroupStart failure.");
}
row_comm.group_start();
for (size_t i = 0; i < num_concurrent_bcasts; ++i) {
auto j = num_rounds * i + round;
if (j < static_cast<size_t>(row_comm_size)) {
Expand All @@ -343,10 +338,8 @@ void update_edge_partition_minor_property(
handle.get_stream());
}
}
// FIXME: better use a higher-level interface than directly invoking NCCL functions.
if (ncclGroupEnd() != ncclSuccess) {
CUGRAPH_FAIL("ncclGroupEnd failure.");
}
row_comm.group_end();

for (size_t i = 0; i < num_concurrent_bcasts; ++i) {
auto j = num_rounds * i + round;
if (j < static_cast<size_t>(row_comm_size)) {
Expand All @@ -363,33 +356,6 @@ void update_edge_partition_minor_property(
}
}
}
#else
vertex_t max_rx_size{0};
for (int i = 0; i < row_comm_size; ++i) {
max_rx_size = std::max(
max_rx_size, graph_view.vertex_partition_range_size(col_comm_rank * row_comm_size + i));
}
auto rx_value_buffer = allocate_dataframe_buffer<value_t>(max_rx_size, handle.get_stream());
auto rx_value_first = get_dataframe_buffer_begin(rx_value_buffer);
for (int i = 0; i < row_comm_size; ++i) {
device_bcast(row_comm,
vertex_property_input_first,
rx_value_first,
graph_view.vertex_partition_range_size(col_comm_rank * row_comm_size + i),
i,
handle.get_stream());

auto v_offset_first = thrust::make_transform_iterator(
(*(edge_partition_minor_property_output.keys())).begin() + key_offsets[i],
[v_first = graph_view.vertex_partition_range_first(
col_comm_rank * row_comm_size + i)] __device__(auto v) { return v - v_first; });
thrust::gather(handle.get_thrust_policy(),
v_offset_first,
v_offset_first + (key_offsets[i + 1] - key_offsets[i]),
rx_value_first,
edge_partition_minor_property_output.value_first() + key_offsets[i]);
}
#endif
} else {
std::vector<size_t> rx_counts(row_comm_size, size_t{0});
std::vector<size_t> displacements(row_comm_size, size_t{0});
Expand Down