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

Fixed the performance regression within EvaluateSplits(). #3680

Merged
merged 1 commit into from
Sep 8, 2018
Merged

Fixed the performance regression within EvaluateSplits(). #3680

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
14 changes: 6 additions & 8 deletions src/tree/updater_gpu_hist.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -747,6 +747,7 @@ class GPUHistMaker : public TreeUpdater {
struct ExpandEntry;

GPUHistMaker() : initialised_(false), p_last_fmat_(nullptr) {}

void Init(
const std::vector<std::pair<std::string, std::string>>& args) override {
param_.InitAllowUnknown(args);
Expand Down Expand Up @@ -919,9 +920,7 @@ class GPUHistMaker : public TreeUpdater {
const std::vector<int>& nidx_set, RegTree* p_tree) {
auto columns = info_->num_col_;
std::vector<DeviceSplitCandidate> best_splits(nidx_set.size());
DeviceSplitCandidate* candidate_splits;
dh::safe_cuda(cudaMallocHost(&candidate_splits, nidx_set.size() *
columns * sizeof(DeviceSplitCandidate)));
std::vector<DeviceSplitCandidate> candidate_splits(nidx_set.size() * columns);
// Use first device
auto& shard = shards_.front();
dh::safe_cuda(cudaSetDevice(shard->device_idx));
Expand Down Expand Up @@ -952,10 +951,10 @@ class GPUHistMaker : public TreeUpdater {
}

dh::safe_cuda(cudaDeviceSynchronize());
dh::safe_cuda(
cudaMemcpy(candidate_splits, shard->temp_memory.d_temp_storage,
sizeof(DeviceSplitCandidate) * columns * nidx_set.size(),
cudaMemcpyDeviceToHost));
dh::safe_cuda
(cudaMemcpy(candidate_splits.data(), shard->temp_memory.d_temp_storage,
sizeof(DeviceSplitCandidate) * columns * nidx_set.size(),
cudaMemcpyDeviceToHost));
for (auto i = 0; i < nidx_set.size(); i++) {
auto depth = p_tree->GetDepth(nidx_set[i]);
DeviceSplitCandidate nidx_best;
Expand All @@ -965,7 +964,6 @@ class GPUHistMaker : public TreeUpdater {
}
best_splits[i] = nidx_best;
}
dh::safe_cuda(cudaFreeHost(candidate_splits));
return std::move(best_splits);
}

Expand Down