Skip to content

Topology fix - tree depth and lint fix #8

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
thatguymike wants to merge 2 commits into
base: parallel
Choose a base branch
from
Open

Topology fix - tree depth and lint fix #8

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
72dc660
Fix outside loop to loop for full tree depth
Jul 17, 2015
c21420c
Fix lint errors
Jul 20, 2015
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
29 changes: 17 additions & 12 deletions src/caffe/parallel.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -119,8 +119,11 @@ void DevicePair::compute(const vector<int> devices, vector<DevicePair>* pairs) {
#ifndef CPU_ONLY
vector<int> remaining(devices);

// Group GPUs by board - some boards can have more than 2 ASICs
for (int d = 0; d < remaining.size(); ++d) {
// Depth for reduction tree
int remaining_depth = static_cast<int>(ceil(log2(remaining.size())));

// Group GPUs by board
for (int d = 0; d < remaining_depth; ++d) {
for (int i = 0; i < remaining.size(); ++i) {
for (int j = i + 1; j < remaining.size(); ++j) {
cudaDeviceProp a, b;
Expand All @@ -144,8 +147,9 @@ void DevicePair::compute(const vector<int> devices, vector<DevicePair>* pairs) {
}
DLOG(INFO) << "GPUs paired by boards, remaining: " << s.str();

// Group by P2P accessibility - P2P group can be larger than 4 boards
for (int d = 0; d < remaining.size(); ++d) {
// Group by P2P accessibility
remaining_depth = ceil(log2(remaining.size()));
for (int d = 0; d < remaining_depth; ++d) {
for (int i = 0; i < remaining.size(); ++i) {
for (int j = i + 1; j < remaining.size(); ++j) {
int access;
Expand All @@ -169,18 +173,19 @@ void DevicePair::compute(const vector<int> devices, vector<DevicePair>* pairs) {
DLOG(INFO) << "GPUs paired by P2P access, remaining: " << s.str();

// Group remaining
for (int d = 0; d < remaining.size(); ++d) { // try to pair everyone
for (int i = 0; i < remaining.size(); ++i) {
for (int j = i + 1; j < remaining.size(); ++j) {
pairs->push_back(DevicePair(remaining[i], remaining[j]));
remaining_depth = ceil(log2(remaining.size()));
for (int d = 0; d < remaining_depth; ++d) {
for (int i = 0; i < remaining.size(); ++i) {
pairs->push_back(DevicePair(remaining[i], remaining[i+1]));
DLOG(INFO) << "Remaining pair: " << remaining[i]
<< ":" << remaining[j];
remaining.erase(remaining.begin() + j);
break;
}
<< ":" << remaining[i+1];
remaining.erase(remaining.begin() + i+1);
}
}

// Should only be the parent node remaining
CHECK_EQ(remaining.size(), 1);

pairs->insert(pairs->begin(), DevicePair(-1, remaining[0]));

CHECK(pairs->size() == devices.size());
Expand Down