Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix for issue 3306. (#3324)

Browse Source
This commit is contained in:
Thejaswi
2018-05-23 07:12:20 +05:30
committed by Rory Mitchell
parent 8f6aadd4b7
commit d367e4fc6b
3 changed files with 6 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/objective/regression_obj.cc
View File

@@ -55,9 +55,7 @@ class RegLossObj : public ObjFunction {
avx::Float8 scale(param_.scale_pos_weight);
const omp_ulong remainder = n % 8;
int nthread = omp_get_max_threads();
// Use a maximum of 8 threads
#pragma omp parallel for schedule(static) num_threads(std::min(8, nthread))
#pragma omp parallel for schedule(static)
for (omp_ulong i = 0; i < n - remainder; i += 8) {
avx::Float8 y(&info.labels_[i]);
avx::Float8 p = Loss::PredTransform(avx::Float8(&preds_h[i]));
@@ -77,9 +75,6 @@ class RegLossObj : public ObjFunction {
gpair[i] = GradientPair(Loss::FirstOrderGradient(p, y) * w,
Loss::SecondOrderGradient(p, y) * w);
}
// Reset omp max threads
omp_set_num_threads(nthread);
}
const char *DefaultEvalMetric() const override {
return Loss::DefaultEvalMetric();

4
src/objective/regression_obj_gpu.cu
View File

@@ -136,7 +136,7 @@ class GPURegLossObj : public ObjFunction {
}
// run the kernel
#pragma omp parallel for schedule(static, 1)
#pragma omp parallel for schedule(static, 1) if (devices_.Size() > 1)
for (int i = 0; i < devices_.Size(); ++i) {
int d = devices_[i];
dh::safe_cuda(cudaSetDevice(d));
@@ -173,7 +173,7 @@ class GPURegLossObj : public ObjFunction {
}
void PredTransformDevice(HostDeviceVector<float>* preds) {
#pragma omp parallel for schedule(static, 1)
#pragma omp parallel for schedule(static, 1) if (devices_.Size() > 1)
for (int i = 0; i < devices_.Size(); ++i) {
int d = devices_[i];
dh::safe_cuda(cudaSetDevice(d));