From 11163dfc7a6c1504c42acf77bd48d677d7167941 Mon Sep 17 00:00:00 2001 From: qijun Date: Wed, 6 Sep 2017 11:54:33 +0800 Subject: [PATCH] make dim int to int64_t --- paddle/framework/ddim.cc | 64 ++++++++++----------- paddle/framework/ddim.h | 20 +++---- paddle/framework/ddim_test.cc | 4 +- paddle/framework/dim.h | 67 +++++++++++----------- paddle/framework/dim_test.cu | 4 +- paddle/framework/eigen.h | 2 +- paddle/framework/tensor_impl.h | 4 +- paddle/operators/gaussian_random_op.cc | 10 +++- paddle/operators/rnn/recurrent_op_utils.cc | 2 +- paddle/operators/uniform_random_op.cc | 10 +++- paddle/pybind/pybind.cc | 2 +- paddle/pybind/tensor_py.h | 4 +- 12 files changed, 102 insertions(+), 91 deletions(-) diff --git a/paddle/framework/ddim.cc b/paddle/framework/ddim.cc index cfd3e8dfde..85b7de7974 100644 --- a/paddle/framework/ddim.cc +++ b/paddle/framework/ddim.cc @@ -21,16 +21,16 @@ namespace framework { /// @cond HIDDEN template -Dim make_dim(const int* d) { +Dim make_dim(const int64_t* d) { return Dim(*d, make_dim(d + 1)); } template <> -Dim<1> make_dim<1>(const int* d) { +Dim<1> make_dim<1>(const int64_t* d) { return Dim<1>(*d); } -void make_ddim(DDim& ddim, const int* dims, int n) { +void make_ddim(DDim& ddim, const int64_t* dims, int n) { switch (n) { case 1: ddim = make_dim<1>(dims); @@ -67,13 +67,13 @@ void make_ddim(DDim& ddim, const int* dims, int n) { /// @endcond -DDim make_ddim(std::initializer_list dims) { +DDim make_ddim(std::initializer_list dims) { DDim result(make_dim(0)); make_ddim(result, dims.begin(), dims.size()); return result; } -DDim make_ddim(const std::vector& dims) { +DDim make_ddim(const std::vector& dims) { DDim result(make_dim(0)); make_ddim(result, &dims[0], dims.size()); return result; @@ -81,12 +81,12 @@ DDim make_ddim(const std::vector& dims) { /// @cond HIDDEN // XXX For some reason, putting this in an anonymous namespace causes errors -class DynamicMutableIndexer : public boost::static_visitor { +class DynamicMutableIndexer : public boost::static_visitor { public: explicit DynamicMutableIndexer(int idx) : idx_(idx) {} template - int& operator()(Dim& dim) const { + int64_t& operator()(Dim& dim) const { return dim[idx_]; } @@ -94,12 +94,12 @@ class DynamicMutableIndexer : public boost::static_visitor { int idx_; }; -class DynamicConstIndexer : public boost::static_visitor { +class DynamicConstIndexer : public boost::static_visitor { public: explicit DynamicConstIndexer(int idx) : idx_(idx) {} template - int operator()(const Dim& dim) const { + int64_t operator()(const Dim& dim) const { return dim[idx_]; } @@ -109,22 +109,22 @@ class DynamicConstIndexer : public boost::static_visitor { /// @endcond -int& DDim::operator[](int idx) { +int64_t& DDim::operator[](int idx) { return boost::apply_visitor(DynamicMutableIndexer(idx), var); } -int DDim::operator[](int idx) const { +int64_t DDim::operator[](int idx) const { return boost::apply_visitor(DynamicConstIndexer(idx), var); } -ssize_t DDim::size() const { return arity(*this); } +int64_t DDim::size() const { return arity(*this); } bool DDim::operator==(DDim d) const { if (var.which() != d.getVar().which()) { return false; } else { - std::vector v1 = vectorize(*this); - std::vector v2 = vectorize(d); + std::vector v1 = vectorize(*this); + std::vector v2 = vectorize(d); for (unsigned int i = 0; i < v1.size(); i++) { if (v1[i] != v2[i]) { @@ -139,10 +139,10 @@ bool DDim::operator==(DDim d) const { bool DDim::operator!=(DDim d) const { return !(*this == d); } DDim DDim::operator+(DDim d) const { - std::vector v1 = vectorize(*this); - std::vector v2 = vectorize(d); + std::vector v1 = vectorize(*this); + std::vector v2 = vectorize(d); - std::vector v3; + std::vector v3; assert(v1.size() == v2.size()); @@ -154,10 +154,10 @@ DDim DDim::operator+(DDim d) const { } DDim DDim::operator*(DDim d) const { - std::vector v1 = vectorize(*this); - std::vector v2 = vectorize(d); + std::vector v1 = vectorize(*this); + std::vector v2 = vectorize(d); - std::vector v3; + std::vector v3; assert(v1.size() == v2.size()); @@ -168,15 +168,15 @@ DDim DDim::operator*(DDim d) const { return make_ddim(v3); } -int get(const DDim& ddim, int idx) { return ddim[idx]; } +int64_t get(const DDim& ddim, int idx) { return ddim[idx]; } void set(DDim& ddim, int idx, int value) { ddim[idx] = value; } /// @cond HIDDEN struct VectorizeVisitor : public boost::static_visitor<> { - std::vector& vector; + std::vector& vector; - explicit VectorizeVisitor(std::vector& v) : vector(v) {} + explicit VectorizeVisitor(std::vector& v) : vector(v) {} template void operator()(const T& t) { @@ -188,31 +188,31 @@ struct VectorizeVisitor : public boost::static_visitor<> { }; /// @endcond -std::vector vectorize(const DDim& ddim) { - std::vector result; +std::vector vectorize(const DDim& ddim) { + std::vector result; VectorizeVisitor visitor(result); boost::apply_visitor(visitor, ddim); return result; } -struct ProductVisitor : public boost::static_visitor { +struct ProductVisitor : public boost::static_visitor { template - ssize_t operator()(const Dim& dim) { + int64_t operator()(const Dim& dim) { return product(dim); } }; -ssize_t product(const DDim& ddim) { +int64_t product(const DDim& ddim) { ProductVisitor visitor; return boost::apply_visitor(visitor, ddim); } struct SliceVectorizeVisitor : public boost::static_visitor<> { - std::vector& vector; + std::vector& vector; int begin; int end; - SliceVectorizeVisitor(std::vector& v, int b, int e) + SliceVectorizeVisitor(std::vector& v, int b, int e) : vector(v), begin(b), end(e) { PADDLE_ENFORCE(begin < end, "Begin index must be less than end index in ddim slice."); @@ -240,7 +240,7 @@ struct SliceVectorizeVisitor : public boost::static_visitor<> { }; DDim slice_ddim(const DDim& dim, int begin, int end) { - std::vector vec; + std::vector vec; vec.reserve(end - begin); SliceVectorizeVisitor visitor(vec, begin, end); boost::apply_visitor(visitor, dim); @@ -280,7 +280,7 @@ std::ostream& operator<<(std::ostream& os, const DDim& ddim) { return os; } -DDim::DDim(std::initializer_list init_list) { +DDim::DDim(std::initializer_list init_list) { *this = make_ddim(init_list); } } // namespace framework diff --git a/paddle/framework/ddim.h b/paddle/framework/ddim.h index 95f294b627..db30c52394 100644 --- a/paddle/framework/ddim.h +++ b/paddle/framework/ddim.h @@ -40,7 +40,7 @@ struct DDim { template explicit DDim(const Dim& in) : var(in) {} - /*implicit*/ DDim(std::initializer_list init_list); + /*implicit*/ DDim(std::initializer_list init_list); template DDim& operator=(const Dim& in) { @@ -48,8 +48,8 @@ struct DDim { return *this; } - int& operator[](int idx); - int operator[](int idx) const; + int64_t& operator[](int idx); + int64_t operator[](int idx) const; template typename Visitor::result_type apply_visitor(Visitor& visitor) { @@ -71,15 +71,15 @@ struct DDim { DDim operator*(DDim d) const; - ssize_t size() const; + int64_t size() const; }; /** - * \brief Make a DDim from std::vector + * \brief Make a DDim from std::vector * * \param dims An vector of ints. Must be sized between [1, 9] */ -DDim make_ddim(const std::vector& dims); +DDim make_ddim(const std::vector& dims); /** * \brief Make a DDim from an initializer list @@ -87,14 +87,14 @@ DDim make_ddim(const std::vector& dims); * \param dims An initializer list of ints. Must be sized between [1, 9] * */ -DDim make_ddim(std::initializer_list dims); +DDim make_ddim(std::initializer_list dims); -int get(const DDim& dim, int idx); +int64_t get(const DDim& dim, int idx); void set(DDim& dim, int idx, int val); -std::vector vectorize(const DDim& ddim); +std::vector vectorize(const DDim& ddim); -ssize_t product(const DDim& ddim); +int64_t product(const DDim& ddim); /** * \brief Slice a ddim diff --git a/paddle/framework/ddim_test.cc b/paddle/framework/ddim_test.cc index 9d18a2972c..756232b1b5 100644 --- a/paddle/framework/ddim_test.cc +++ b/paddle/framework/ddim_test.cc @@ -12,7 +12,7 @@ TEST(DDim, Equality) { EXPECT_EQ(ddim[2], 5); // construct a DDim from a vector - std::vector vec({9, 1, 5}); + std::vector vec({9, 1, 5}); paddle::framework::DDim vddim = paddle::framework::make_ddim(vec); EXPECT_EQ(ddim[0], 9); EXPECT_EQ(ddim[1], 1); @@ -25,7 +25,7 @@ TEST(DDim, Equality) { EXPECT_EQ(paddle::framework::get(ddim, 0), 6); // vectorize a DDim - std::vector res_vec = paddle::framework::vectorize(vddim); + std::vector res_vec = paddle::framework::vectorize(vddim); EXPECT_EQ(res_vec[0], 9); EXPECT_EQ(res_vec[1], 1); EXPECT_EQ(res_vec[2], 5); diff --git a/paddle/framework/dim.h b/paddle/framework/dim.h index 883fdc55eb..04d4b0e604 100644 --- a/paddle/framework/dim.h +++ b/paddle/framework/dim.h @@ -17,13 +17,13 @@ struct Dim { static constexpr int dimensions = i; template - HOSTDEVICE Dim(int _head, Args... _tail) : head(_head), tail(_tail...) { + HOSTDEVICE Dim(int64_t _head, Args... _tail) : head(_head), tail(_tail...) { static_assert(sizeof...(_tail) == i - 1, "Dim initialized with the wrong number of parameters"); } HOSTDEVICE - Dim(int _head, const Dim& _tail) : head(_head), tail(_tail) {} + Dim(int64_t _head, const Dim& _tail) : head(_head), tail(_tail) {} HOSTDEVICE Dim() : head(0), tail() {} @@ -31,12 +31,12 @@ struct Dim { /** Construct a Dim from a linear index and size. Uses Fortran order * indexing. */ HOSTDEVICE - Dim(int idx, const Dim& size) + Dim(int64_t idx, const Dim& size) : head(idx % size.head), tail(idx / size.head, size.tail) {} /** Construct a Dim with each dimension set to the given index */ HOSTDEVICE - Dim(int idx) : head(idx), tail(idx) {} + Dim(int64_t idx) : head(idx), tail(idx) {} HOSTDEVICE bool operator==(const Dim& o) const { @@ -47,13 +47,13 @@ struct Dim { bool operator!=(const Dim& o) const { return !(*this == o); } HOSTDEVICE - int& operator[](int idx); + int64_t& operator[](int idx); HOSTDEVICE - int operator[](int idx) const; + int64_t operator[](int idx) const; HOST std::string to_string() const; - int head; + int64_t head; Dim tail; }; @@ -63,7 +63,7 @@ struct Dim<1> { static constexpr int dimensions = 1; HOSTDEVICE - Dim(int _head) : head(_head) {} + Dim(int64_t _head) : head(_head) {} HOSTDEVICE Dim() : head(0) {} @@ -86,11 +86,11 @@ struct Dim<1> { bool operator!=(const Dim<1>& o) const { return !(*this == o); } HOSTDEVICE - int& operator[](int idx); + int64_t& operator[](int idx); HOSTDEVICE - int operator[](int idx) const; + int64_t operator[](int idx) const; - int head; + int64_t head; }; namespace { @@ -100,12 +100,12 @@ template struct DimGetter { // Return a copy if Dim is const template - HOSTDEVICE static int impl(const D& d) { + HOSTDEVICE static int64_t impl(const D& d) { return DimGetter::impl(d.tail); } // Return a reference if Dim is mutable template - HOSTDEVICE static int& impl(D& d) { + HOSTDEVICE static int64_t& impl(D& d) { return DimGetter::impl(d.tail); } }; @@ -115,18 +115,18 @@ template <> struct DimGetter<0> { // Return a copy if Dim is const template - HOSTDEVICE static int impl(const D& d) { + HOSTDEVICE static int64_t impl(const D& d) { return d.head; } // Return a reference if Dim is mutable template - HOSTDEVICE static int& impl(D& d) { + HOSTDEVICE static int64_t& impl(D& d) { return d.head; } }; template -HOSTDEVICE int& indexer(Dim& dim, int idx) { +HOSTDEVICE int64_t& indexer(Dim& dim, int idx) { #ifndef __CUDA_ARCH__ if (idx < 0) { throw std::invalid_argument("Tried to access a negative dimension"); @@ -141,7 +141,7 @@ HOSTDEVICE int& indexer(Dim& dim, int idx) { } template <> -HOSTDEVICE int& indexer<1>(Dim<1>& dim, int idx) { +HOSTDEVICE int64_t& indexer<1>(Dim<1>& dim, int idx) { #ifndef __CUDA_ARCH__ if (idx != 0) { throw std::invalid_argument("Invalid index"); @@ -153,7 +153,7 @@ HOSTDEVICE int& indexer<1>(Dim<1>& dim, int idx) { } template -HOSTDEVICE int indexer(const Dim& dim, int idx) { +HOSTDEVICE int64_t indexer(const Dim& dim, int idx) { #ifndef __CUDA_ARCH__ if (idx < 0) { throw std::invalid_argument("Tried to access a negative dimension"); @@ -168,7 +168,7 @@ HOSTDEVICE int indexer(const Dim& dim, int idx) { } template <> -HOSTDEVICE int indexer<1>(const Dim<1>& dim, int idx) { +HOSTDEVICE int64_t indexer<1>(const Dim<1>& dim, int idx) { #ifndef __CUDA_ARCH__ if (idx != 0) { throw std::invalid_argument("Invalid index"); @@ -182,73 +182,76 @@ HOSTDEVICE int indexer<1>(const Dim<1>& dim, int idx) { } // namespace // Static access to constant Dim template -HOSTDEVICE int get(const Dim& d) { +HOSTDEVICE int64_t get(const Dim& d) { return DimGetter::impl(d); } // Static access to mutable Dim template -HOSTDEVICE int& get(Dim& d) { +HOSTDEVICE int64_t& get(Dim& d) { return DimGetter::impl(d); } // Dynamic access to constant Dim template -HOSTDEVICE int Dim::operator[](int i) const { +HOSTDEVICE int64_t Dim::operator[](int i) const { return indexer(*this, i); } // Dynamic access to mutable Dim template -HOSTDEVICE int& Dim::operator[](int i) { +HOSTDEVICE int64_t& Dim::operator[](int i) { return indexer(*this, i); } // Dynamic access to constant Dim -inline HOSTDEVICE int Dim<1>::operator[](int i) const { +inline HOSTDEVICE int64_t Dim<1>::operator[](int i) const { return indexer(*this, i); } // Dynamic access to mutable Dim -inline HOSTDEVICE int& Dim<1>::operator[](int i) { return indexer(*this, i); } +inline HOSTDEVICE int64_t& Dim<1>::operator[](int i) { + return indexer(*this, i); +} // Dynamic access to constant Dim // without std::enable_if will try to instantiate this on get<0>(d) template -HOSTDEVICE typename std::enable_if<(l > 0), int>::type get(const Dim& d, - int i) { +HOSTDEVICE typename std::enable_if<(l > 0), int64_t>::type get(const Dim& d, + int i) { return d[i]; } // Dynamic access to mutable Dim template -HOSTDEVICE typename std::enable_if<(l > 0), int&>::type get(Dim& d, int i) { +HOSTDEVICE typename std::enable_if<(l > 0), int64_t&>::type get(Dim& d, + int i) { return d[i]; } // Dot product of two dims template -HOSTDEVICE int linearize(const Dim& a, const Dim& b) { +HOSTDEVICE int64_t linearize(const Dim& a, const Dim& b) { return a.head * b.head + linearize(a.tail, b.tail); } // Base case dot product of two Dims // Notice it is inline because it is no longer a template template <> -HOSTDEVICE inline int linearize(const Dim<1>& a, const Dim<1>& b) { +HOSTDEVICE inline int64_t linearize(const Dim<1>& a, const Dim<1>& b) { return a.head * b.head; } // Product of a Dim template -HOSTDEVICE int product(const Dim& a, int prod = 1) { +HOSTDEVICE int64_t product(const Dim& a, int prod = 1) { return prod * a.head * product(a.tail); } // Base case product of a Dim // Notice it is inline because it is no longer a template template <> -HOSTDEVICE inline int product(const Dim<1>& a, int prod) { +HOSTDEVICE inline int64_t product(const Dim<1>& a, int prod) { return prod * a.head; } diff --git a/paddle/framework/dim_test.cu b/paddle/framework/dim_test.cu index 3898d0a447..f42b1eeadf 100644 --- a/paddle/framework/dim_test.cu +++ b/paddle/framework/dim_test.cu @@ -47,9 +47,9 @@ TEST(Dim, Equality) { EXPECT_EQ(b[1], 11); // dynamic access on GPU - thrust::device_vector r(1); + thrust::device_vector r(1); dyn_idx_gpu<<<1, 1>>>(thrust::raw_pointer_cast(r.data())); - int res = r[0]; + int64_t res = r[0]; EXPECT_EQ(res, 6); // ex_prefix_mul diff --git a/paddle/framework/eigen.h b/paddle/framework/eigen.h index a4667cc51f..2d8d9ae10c 100644 --- a/paddle/framework/eigen.h +++ b/paddle/framework/eigen.h @@ -28,7 +28,7 @@ struct EigenDim { static Type From(const DDim& dims) { PADDLE_ENFORCE(arity(dims) == D, "D must match arity(DDim)"); Type ret; - for (int d = 0; d < arity(dims); d++) { + for (int64_t d = 0; d < arity(dims); d++) { ret[d] = dims[d]; } return ret; diff --git a/paddle/framework/tensor_impl.h b/paddle/framework/tensor_impl.h index 7893e233b7..94f436294f 100644 --- a/paddle/framework/tensor_impl.h +++ b/paddle/framework/tensor_impl.h @@ -58,7 +58,7 @@ inline T* Tensor::mutable_data(platform::Place place) { "Tensor's numel must be larger than zero to call " "Tensor::mutable_data. Call Tensor::set_dim first."); /* some versions of boost::variant don't have operator!= */ - size_t size = product(dims_) * sizeof(T); + int64_t size = product(dims_) * sizeof(T); if (holder_ == nullptr || !(holder_->place() == place) || holder_->size() < size + offset_) { if (platform::is_cpu_place(place)) { @@ -131,7 +131,7 @@ inline Tensor Tensor::Slice(const int& begin_idx, const int& end_idx) const { PADDLE_ENFORCE_LT(begin_idx, end_idx, "Begin index must be less than end index."); PADDLE_ENFORCE_NE(dims_[0], 1, "Can not slice a tensor with dims_[0] = 1."); - int base = product(dims_) / dims_[0]; + size_t base = product(dims_) / dims_[0]; Tensor dst; dst.holder_ = holder_; DDim dst_dims = dims_; diff --git a/paddle/operators/gaussian_random_op.cc b/paddle/operators/gaussian_random_op.cc index 056447901d..583c9e5774 100644 --- a/paddle/operators/gaussian_random_op.cc +++ b/paddle/operators/gaussian_random_op.cc @@ -31,8 +31,8 @@ class CPUGaussianRandomKernel : public framework::OpKernel { } engine.seed(seed); std::normal_distribution dist(mean, std); - ssize_t size = framework::product(tensor->dims()); - for (ssize_t i = 0; i < size; ++i) { + int64_t size = framework::product(tensor->dims()); + for (int64_t i = 0; i < size; ++i) { data[i] = dist(engine); } } @@ -46,9 +46,13 @@ class GaussianRandomOp : public framework::OperatorWithKernel { void InferShape(const framework::InferShapeContext& context) const override { auto* tensor = context.Output("Out"); auto dims = GetAttr>("dims"); + std::vector temp(dims.size()); + for (auto dim : dims) { + temp.push_back(static_cast(dim)); + } PADDLE_ENFORCE(dims.size() > 0UL, "dims can be one int or array. dims must be set."); - tensor->Resize(framework::make_ddim(dims)); + tensor->Resize(framework::make_ddim(temp)); } }; diff --git a/paddle/operators/rnn/recurrent_op_utils.cc b/paddle/operators/rnn/recurrent_op_utils.cc index a9b65c30f2..69e723b401 100644 --- a/paddle/operators/rnn/recurrent_op_utils.cc +++ b/paddle/operators/rnn/recurrent_op_utils.cc @@ -61,7 +61,7 @@ void ConcatOutputs(const std::vector& step_scopes, PADDLE_ENFORCE(step_scope_var != nullptr, "%s not in scope", outlinks[i].internal); f::DDim step_dims = step_scope_var->template GetMutable()->dims(); - std::vector dims_vec = vectorize(step_dims); + std::vector dims_vec = vectorize(step_dims); dims_vec.insert(dims_vec.begin(), seq_len); output->Resize(f::make_ddim(dims_vec)); } else { diff --git a/paddle/operators/uniform_random_op.cc b/paddle/operators/uniform_random_op.cc index 2d943c4508..3241d4289a 100644 --- a/paddle/operators/uniform_random_op.cc +++ b/paddle/operators/uniform_random_op.cc @@ -35,8 +35,8 @@ class CPUUniformRandomKernel : public framework::OpKernel { std::uniform_real_distribution dist( static_cast(context.GetAttr("min")), static_cast(context.GetAttr("max"))); - ssize_t size = framework::product(tensor->dims()); - for (ssize_t i = 0; i < size; ++i) { + int64_t size = framework::product(tensor->dims()); + for (int64_t i = 0; i < size; ++i) { data[i] = dist(engine); } } @@ -52,7 +52,11 @@ class UniformRandomOp : public framework::OperatorWithKernel { "uniform_random's min must less then max"); auto* tensor = ctx.Output("Out"); auto dims = GetAttr>("dims"); - tensor->Resize(framework::make_ddim(dims)); + std::vector temp(dims.size()); + for (auto dim : dims) { + temp.push_back(static_cast(dim)); + } + tensor->Resize(framework::make_ddim(temp)); } }; diff --git a/paddle/pybind/pybind.cc b/paddle/pybind/pybind.cc index 6896422617..7db629a68f 100644 --- a/paddle/pybind/pybind.cc +++ b/paddle/pybind/pybind.cc @@ -76,7 +76,7 @@ PYBIND11_PLUGIN(core) { .def("get_dims", [](const Tensor &self) { return vectorize(self.dims()); }) .def("set_dims", - [](Tensor &self, const std::vector &dim) { + [](Tensor &self, const std::vector &dim) { self.Resize(make_ddim(dim)); }) .def("alloc_float", diff --git a/paddle/pybind/tensor_py.h b/paddle/pybind/tensor_py.h index 39ba60b4dc..95171acf72 100644 --- a/paddle/pybind/tensor_py.h +++ b/paddle/pybind/tensor_py.h @@ -85,7 +85,7 @@ void PyCPUTensorSetFromArray( framework::Tensor &self, py::array_t array, paddle::platform::CPUPlace &place) { - std::vector dims; + std::vector dims; dims.reserve(array.ndim()); for (size_t i = 0; i < array.ndim(); ++i) { dims.push_back((int)array.shape()[i]); @@ -102,7 +102,7 @@ void PyCUDATensorSetFromArray( framework::Tensor &self, py::array_t array, paddle::platform::GPUPlace &place) { - std::vector dims; + std::vector dims; dims.reserve(array.ndim()); for (size_t i = 0; i < array.ndim(); ++i) { dims.push_back((int)array.shape()[i]); -- GitLab