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

Make HostDeviceVector single gpu only (#4773)

Browse Source 
* Make HostDeviceVector single gpu only
This commit is contained in:
Rong Ou
2019-08-25 14:51:13 -07:00
committed by Rory Mitchell
parent 41227d1933
commit 38ab79f889
54 changed files with 641 additions and 1621 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/objective/hinge.cu
View File

@@ -58,7 +58,7 @@ class HingeObj : public ObjFunction {
_out_gpair[_idx] = GradientPair(g, h);
},
common::Range{0, static_cast<int64_t>(ndata)},
GPUSet::All(tparam_->gpu_id, tparam_->n_gpus, ndata)).Eval(
tparam_->gpu_id).Eval(
out_gpair, &preds, &info.labels_, &info.weights_);
}
@@ -68,7 +68,7 @@ class HingeObj : public ObjFunction {
_preds[_idx] = _preds[_idx] > 0.0 ? 1.0 : 0.0;
},
common::Range{0, static_cast<int64_t>(io_preds->Size()), 1},
GPUSet::All(tparam_->gpu_id, tparam_->n_gpus, io_preds->Size()))
tparam_->gpu_id)
.Eval(io_preds);
}

26
src/objective/multiclass_obj.cu
View File

@@ -59,14 +59,14 @@ class SoftmaxMultiClassObj : public ObjFunction {
const int nclass = param_.num_class;
const auto ndata = static_cast<int64_t>(preds.Size() / nclass);
auto devices = GPUSet::All(tparam_->gpu_id, tparam_->n_gpus, preds.Size());
out_gpair->Shard(GPUDistribution::Granular(devices, nclass));
info.labels_.Shard(GPUDistribution::Block(devices));
info.weights_.Shard(GPUDistribution::Block(devices));
preds.Shard(GPUDistribution::Granular(devices, nclass));
auto device = tparam_->gpu_id;
out_gpair->SetDevice(device);
info.labels_.SetDevice(device);
info.weights_.SetDevice(device);
preds.SetDevice(device);
label_correct_.Resize(devices.IsEmpty() ? 1 : devices.Size());
label_correct_.Shard(GPUDistribution::Block(devices));
label_correct_.Resize(1);
label_correct_.SetDevice(device);
out_gpair->Resize(preds.Size());
label_correct_.Fill(1);
@@ -100,7 +100,7 @@ class SoftmaxMultiClassObj : public ObjFunction {
p = label == k ? p - 1.0f : p;
gpair[idx * nclass + k] = GradientPair(p * wt, h);
}
}, common::Range{0, ndata}, devices, false)
}, common::Range{0, ndata}, device, false)
.Eval(out_gpair, &info.labels_, &preds, &info.weights_, &label_correct_);
std::vector<int>& label_correct_h = label_correct_.HostVector();
@@ -125,7 +125,7 @@ class SoftmaxMultiClassObj : public ObjFunction {
const auto ndata = static_cast<int64_t>(io_preds->Size() / nclass);
max_preds_.Resize(ndata);
auto devices = GPUSet::All(tparam_->gpu_id, tparam_->n_gpus, io_preds->Size());
auto device = tparam_->gpu_id;
if (prob) {
common::Transform<>::Init(
[=] XGBOOST_DEVICE(size_t _idx, common::Span<bst_float> _preds) {
@@ -133,11 +133,11 @@ class SoftmaxMultiClassObj : public ObjFunction {
_preds.subspan(_idx * nclass, nclass);
common::Softmax(point.begin(), point.end());
},
common::Range{0, ndata}, GPUDistribution::Granular(devices, nclass))
common::Range{0, ndata}, device)
.Eval(io_preds);
} else {
io_preds->Shard(GPUDistribution::Granular(devices, nclass));
max_preds_.Shard(GPUDistribution::Block(devices));
io_preds->SetDevice(device);
max_preds_.SetDevice(device);
common::Transform<>::Init(
[=] XGBOOST_DEVICE(size_t _idx,
common::Span<const bst_float> _preds,
@@ -148,7 +148,7 @@ class SoftmaxMultiClassObj : public ObjFunction {
common::FindMaxIndex(point.cbegin(),
point.cend()) - point.cbegin();
},
common::Range{0, ndata}, devices, false)
common::Range{0, ndata}, device, false)
.Eval(io_preds, &max_preds_);
}
if (!prob) {

32
src/objective/regression_obj.cu
View File

@@ -57,8 +57,8 @@ class RegLossObj : public ObjFunction {
<< "preds.size=" << preds.Size() << ", label.size=" << info.labels_.Size();
size_t ndata = preds.Size();
out_gpair->Resize(ndata);
auto devices = GPUSet::All(tparam_->gpu_id, tparam_->n_gpus, preds.Size());
label_correct_.Resize(devices.IsEmpty() ? 1 : devices.Size());
auto device = tparam_->gpu_id;
label_correct_.Resize(1);
label_correct_.Fill(1);
bool is_null_weight = info.weights_.Size() == 0;
@@ -83,7 +83,7 @@ class RegLossObj : public ObjFunction {
_out_gpair[_idx] = GradientPair(Loss::FirstOrderGradient(p, label) * w,
Loss::SecondOrderGradient(p, label) * w);
},
common::Range{0, static_cast<int64_t>(ndata)}, devices).Eval(
common::Range{0, static_cast<int64_t>(ndata)}, device).Eval(
&label_correct_, out_gpair, &preds, &info.labels_, &info.weights_);
// copy "label correct" flags back to host
@@ -105,7 +105,7 @@ class RegLossObj : public ObjFunction {
[] XGBOOST_DEVICE(size_t _idx, common::Span<float> _preds) {
_preds[_idx] = Loss::PredTransform(_preds[_idx]);
}, common::Range{0, static_cast<int64_t>(io_preds->Size())},
GPUSet::All(tparam_->gpu_id, tparam_->n_gpus, io_preds->Size()))
tparam_->gpu_id)
.Eval(io_preds);
}
@@ -175,8 +175,8 @@ class PoissonRegression : public ObjFunction {
CHECK_EQ(preds.Size(), info.labels_.Size()) << "labels are not correctly provided";
size_t ndata = preds.Size();
out_gpair->Resize(ndata);
auto devices = GPUSet::All(tparam_->gpu_id, tparam_->n_gpus, preds.Size());
label_correct_.Resize(devices.IsEmpty() ? 1 : devices.Size());
auto device = tparam_->gpu_id;
label_correct_.Resize(1);
label_correct_.Fill(1);
bool is_null_weight = info.weights_.Size() == 0;
@@ -197,7 +197,7 @@ class PoissonRegression : public ObjFunction {
_out_gpair[_idx] = GradientPair{(expf(p) - y) * w,
expf(p + max_delta_step) * w};
},
common::Range{0, static_cast<int64_t>(ndata)}, devices).Eval(
common::Range{0, static_cast<int64_t>(ndata)}, device).Eval(
&label_correct_, out_gpair, &preds, &info.labels_, &info.weights_);
// copy "label correct" flags back to host
std::vector<int>& label_correct_h = label_correct_.HostVector();
@@ -213,7 +213,7 @@ class PoissonRegression : public ObjFunction {
_preds[_idx] = expf(_preds[_idx]);
},
common::Range{0, static_cast<int64_t>(io_preds->Size())},
GPUSet::All(tparam_->gpu_id, tparam_->n_gpus, io_preds->Size()))
tparam_->gpu_id)
.Eval(io_preds);
}
void EvalTransform(HostDeviceVector<bst_float> *io_preds) override {
@@ -340,9 +340,9 @@ class GammaRegression : public ObjFunction {
CHECK_NE(info.labels_.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels_.Size()) << "labels are not correctly provided";
const size_t ndata = preds.Size();
auto devices = GPUSet::All(tparam_->gpu_id, tparam_->n_gpus, ndata);
auto device = tparam_->gpu_id;
out_gpair->Resize(ndata);
label_correct_.Resize(devices.IsEmpty() ? 1 : devices.Size());
label_correct_.Resize(1);
label_correct_.Fill(1);
const bool is_null_weight = info.weights_.Size() == 0;
@@ -361,7 +361,7 @@ class GammaRegression : public ObjFunction {
}
_out_gpair[_idx] = GradientPair((1 - y / expf(p)) * w, y / expf(p) * w);
},
common::Range{0, static_cast<int64_t>(ndata)}, devices).Eval(
common::Range{0, static_cast<int64_t>(ndata)}, device).Eval(
&label_correct_, out_gpair, &preds, &info.labels_, &info.weights_);
// copy "label correct" flags back to host
@@ -378,7 +378,7 @@ class GammaRegression : public ObjFunction {
_preds[_idx] = expf(_preds[_idx]);
},
common::Range{0, static_cast<int64_t>(io_preds->Size())},
GPUSet::All(tparam_->gpu_id, tparam_->n_gpus, io_preds->Size()))
tparam_->gpu_id)
.Eval(io_preds);
}
void EvalTransform(HostDeviceVector<bst_float> *io_preds) override {
@@ -430,8 +430,8 @@ class TweedieRegression : public ObjFunction {
const size_t ndata = preds.Size();
out_gpair->Resize(ndata);
auto devices = GPUSet::All(tparam_->gpu_id, tparam_->n_gpus, preds.Size());
label_correct_.Resize(devices.IsEmpty() ? 1 : devices.Size());
auto device = tparam_->gpu_id;
label_correct_.Resize(1);
label_correct_.Fill(1);
const bool is_null_weight = info.weights_.Size() == 0;
@@ -455,7 +455,7 @@ class TweedieRegression : public ObjFunction {
std::exp((1 - rho) * p) + (2 - rho) * expf((2 - rho) * p);
_out_gpair[_idx] = GradientPair(grad * w, hess * w);
},
common::Range{0, static_cast<int64_t>(ndata), 1}, devices)
common::Range{0, static_cast<int64_t>(ndata), 1}, device)
.Eval(&label_correct_, out_gpair, &preds, &info.labels_, &info.weights_);
// copy "label correct" flags back to host
@@ -472,7 +472,7 @@ class TweedieRegression : public ObjFunction {
_preds[_idx] = expf(_preds[_idx]);
},
common::Range{0, static_cast<int64_t>(io_preds->Size())},
GPUSet::All(tparam_->gpu_id, tparam_->n_gpus, io_preds->Size()))
tparam_->gpu_id)
.Eval(io_preds);
}