Skip to content

GitLab

Commit aebbb122 authored by Benjamin Thomas Graham's avatar Benjamin Thomas Graham
Browse files

Merge branch 'master' of github.com:facebookresearch/SparseConvNet

parents 8c706edd e55e3bea
Show whitespace changes
Inline Side-by-side
Showing with 35 additions and 31 deletions
develop.sh 0 → 100644
View file @ aebbb122
#!/bin/bash
rm -rf build/ dist/ sparseconvnet.egg-info sparseconvnet_SCN*.so
python setup.py develop
python examples/hello-world.py
sparseconvnet/SCN/CUDA/ActivePooling.cu
View file @ aebbb122
......@@ -23,7 +23,7 @@ void ActivePooling_ForwardPass(T *input_features, T *output_features,
Int batchSize, Int maxActive, Int nPlanes,
Int *rules, bool average) {
auto rulesBuffer = at::CUDA(at_kINT).tensor({1 << 22});
auto rulesBuffer = at::empty({1<<22}, at::CUDA(at_kINT));
Int *rb = rulesBuffer.data<Int>();
Int rowBatchSize = std::min((Int)32768, (1 << 22) / (maxActive + 1));
assert(rowBatchSize > 0);
......@@ -58,7 +58,7 @@ template <typename T>
void ActivePooling_BackwardPass(T *d_input_features, T *d_output_features,
Int batchSize, Int maxActive, Int nPlanes,
Int *rules, bool average) {
auto rulesBuffer = at::CUDA(at_kINT).tensor({1 << 22});
auto rulesBuffer = at::empty({1<<22}, at::CUDA(at_kINT));
Int *rb = rulesBuffer.data<Int>();
Int rowBatchSize = std::min((Int)32768, (1 << 22) / (maxActive + 1));
assert(rowBatchSize > 0);
......
sparseconvnet/SCN/CUDA/IOLayers.cpp
View file @ aebbb122
......@@ -33,7 +33,7 @@ void cuda_InputLayer_updateOutput(Metadata<Dimension> &m,
} else {
output_features.resize_({*m.inputNActive, nPlanes});
output_features.zero_();
auto rulesBuffer = at::CUDA(at_kINT).tensor({(int)rules[1].size()});
auto rulesBuffer = at::empty({(int)rules[1].size()}, at::CUDA(at_kINT));
auto iF = input_features.data<T>();
auto oF = output_features.data<T>();
Int *rb = rulesBuffer.data<Int>();
......@@ -58,7 +58,7 @@ void cuda_InputLayer_updateGradInput(
} else {
d_input_features.resize_({rules[0][2], nPlanes});
d_input_features.zero_();
auto rulesBuffer = at::CUDA(at_kINT).tensor({(int)rules[1].size()});
auto rulesBuffer = at::empty({(int)rules[1].size()}, at::CUDA(at_kINT));
auto diF = d_input_features.data<T>();
auto doF = d_output_features.data<T>();
Int *rb = rulesBuffer.data<Int>();
......@@ -83,7 +83,7 @@ void cuda_OutputLayer_updateOutput(Metadata<Dimension> &m,
} else {
output_features.resize_({rules[0][2], nPlanes});
output_features.zero_();
auto rulesBuffer = at::CUDA(at_kINT).tensor({(int)rules[1].size()});
auto rulesBuffer = at::empty({(int)rules[1].size()}, at::CUDA(at_kINT));
auto iF = input_features.data<T>();
auto oF = output_features.data<T>();
Int *rb = rulesBuffer.data<Int>();
......@@ -107,7 +107,7 @@ void cuda_OutputLayer_updateGradInput(
} else {
d_input_features.resize_({nRows, nPlanes});
d_input_features.zero_();
auto rulesBuffer = at::CUDA(at_kINT).tensor({(int)rules[1].size()});
auto rulesBuffer = at::empty({(int)rules[1].size()}, at::CUDA(at_kINT));
auto diF = d_input_features.data<T>();
auto doF = d_output_features.data<T>();
Int *rb = rulesBuffer.data<Int>();
......@@ -137,7 +137,7 @@ void cuda_BLInputLayer_updateOutput(Metadata<Dimension> &m,
output_features.copy_(input_features);
output_features.resize_({*m.inputNActive, nPlanes});
} else {
auto rulesBuffer = at::CUDA(at_kINT).tensor({(int)rules[1].size()});
auto rulesBuffer = at::empty({(int)rules[1].size()}, at::CUDA(at_kINT));
auto iF = input_features.data<T>();
auto oF = output_features.data<T>();
Int *rb = rulesBuffer.data<Int>();
......@@ -164,7 +164,7 @@ void cuda_BLInputLayer_updateGradInput(
} else {
d_input_features.resize_({rules[0][2], rules[0][3], nPlanes});
d_input_features.zero_();
auto rulesBuffer = at::CUDA(at_kINT).tensor({(int)rules[1].size()});
auto rulesBuffer = at::empty({(int)rules[1].size()}, at::CUDA(at_kINT));
auto diF = d_input_features.data<T>();
auto doF = d_output_features.data<T>();
Int *rb = rulesBuffer.data<Int>();
......@@ -191,7 +191,7 @@ void cuda_BLOutputLayer_updateOutput(
} else {
output_features.resize_({rules[0][2], rules[0][3], nPlanes});
output_features.zero_();
auto rulesBuffer = at::CUDA(at_kINT).tensor({(int)rules[1].size()});
auto rulesBuffer = at::empty({(int)rules[1].size()}, at::CUDA(at_kINT));
auto iF = input_features.data<T>();
auto oF = output_features.data<T>();
Int *rb = rulesBuffer.data<Int>();
......@@ -216,7 +216,7 @@ void cuda_BLOutputLayer_updateGradInput(
} else {
d_input_features.resize_({nRows, nPlanes});
d_input_features.zero_();
auto rulesBuffer = at::CUDA(at_kINT).tensor({(int)rules[1].size()});
auto rulesBuffer = at::empty({(int)rules[1].size()}, at::CUDA(at_kINT));
auto diF = d_input_features.data<T>();
auto doF = d_output_features.data<T>();
Int *rb = rulesBuffer.data<Int>();
......
sparseconvnet/SCN/CUDA/RuleBookIterator.h
View file @ aebbb122
......@@ -17,7 +17,7 @@
Int rbMaxSize = 0; \
for (auto &r : _rules) \
rbMaxSize = std::max(rbMaxSize, (Int)r.size()); \
at::Tensor rulesBuffer = at::CUDA(at_kINT).tensor({rbMaxSize}); \
at::Tensor rulesBuffer = at::empty({rbMaxSize}, at::CUDA(at_kINT)); \
Int *rbB = rulesBuffer.data<Int>(); \
for (int k = 0; k < _rules.size(); ++k) { \
auto &r = _rules[k]; \
......
sparseconvnet/SCN/Metadata/Metadata.cpp
View file @ aebbb122
......@@ -588,13 +588,13 @@ Metadata<dimension>::compareSparseHelper(Metadata<dimension> &mR,
}
}
}
at::Tensor cL_ = torch::CPU(at::kLong).tensor({(long)cL.size()});
at::Tensor cL_ = at::empty({(long)cL.size()}, at::CPU(at::kLong));
std::memcpy(cL_.data<long>(), &cL[0], cL.size() * sizeof(long));
at::Tensor cR_ = torch::CPU(at::kLong).tensor({(long)cR.size()});
at::Tensor cR_ = at::empty({(long)cR.size()}, at::CPU(at::kLong));
std::memcpy(cR_.data<long>(), &cR[0], cR.size() * sizeof(long));
at::Tensor L_ = torch::CPU(at::kLong).tensor({(long)L.size()});
at::Tensor L_ = at::empty({(long)L.size()}, at::CPU(at::kLong));
std::memcpy(L_.data<long>(), &L[0], L.size() * sizeof(long));
at::Tensor R_ = torch::CPU(at::kLong).tensor({(long)R.size()});
at::Tensor R_ = at::empty({(long)R.size()}, at::CPU(at::kLong));
std::memcpy(R_.data<long>(), &R[0], R.size() * sizeof(long));
return {cL_, cR_, L_, R_};
}
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment