| #pragma once |
| |
| #include "Generator.hpp" |
| #include "Storage.hpp" |
| #include "Type.hpp" |
| |
| #include <TH/TH.h> |
| #include <cstddef> |
| #include <cstdint> |
| #include <initializer_list> |
| #include <memory> |
| #include <type_traits> |
| #include <unordered_map> |
| #include <vector> |
| #include <cstring> |
| #include <stdexcept> |
| |
| namespace thpp { |
| |
| struct THPP_CLASS Tensor { |
| using long_range = std::vector<int64_t>; |
| |
| Tensor() {}; |
| Tensor(const Tensor& other) = delete; |
| Tensor(Tensor&& other) = delete; |
| virtual ~Tensor() {}; |
| |
| virtual Tensor* clone() const = 0; |
| virtual Tensor* clone_shallow() = 0; |
| virtual std::unique_ptr<Tensor> contiguous() const = 0; |
| virtual Tensor* newSelect(int dimension, int64_t sliceIndex) const = 0; |
| virtual Tensor* newNarrow(int dimension, int64_t firstIndex, int64_t size) const = 0; |
| virtual Tensor* newTranspose(int dimension1, int dimension2) const = 0; |
| virtual Tensor* newUnfold(int dimension, int64_t size, int64_t step) const = 0; |
| virtual Tensor* newExpand(const long_range& size) const = 0; |
| virtual Tensor* newView(const long_range& size) const = 0; |
| |
| virtual int nDim() const = 0; |
| virtual long_range sizes() const = 0; |
| virtual long_range strides() const = 0; |
| virtual const int64_t* rawSizes() const = 0; |
| virtual const int64_t* rawStrides() const = 0; |
| virtual std::size_t storageOffset() const = 0; |
| virtual std::size_t elementSize() const = 0; |
| virtual int64_t numel() const = 0; |
| virtual bool isContiguous() const = 0; |
| virtual void* data() = 0; |
| virtual const void* data() const = 0; |
| virtual void* cdata() = 0; |
| virtual const void* cdata() const = 0; |
| virtual Tensor& retain() = 0; |
| virtual Tensor& free() = 0; |
| |
| virtual Tensor& resize(const std::initializer_list<int64_t>& new_size) = 0; |
| virtual Tensor& resize(const std::vector<int64_t>& new_size) = 0; |
| virtual Tensor& resize(THLongStorage *size, |
| THLongStorage *stride) = 0; |
| virtual Tensor& resizeAs(const Tensor& src) = 0; |
| virtual Tensor& set(const Tensor& src) = 0; |
| virtual Tensor& setStorage(const Storage& storage, |
| ptrdiff_t storageOffset, |
| const long_range& size, |
| const long_range& stride) = 0; |
| virtual Tensor& setStorage(const Storage& storage, |
| ptrdiff_t storageOffset, |
| THLongStorage *size, |
| THLongStorage *stride) = 0; |
| virtual Tensor& narrow(const Tensor& src, |
| int dimension, |
| int64_t firstIndex, |
| int64_t size) = 0; |
| virtual Tensor& select(const Tensor& src, int dimension, int64_t sliceIndex) = 0; |
| virtual Tensor& transpose(const Tensor& src, int dimension1, int dimension2) = 0; |
| virtual Tensor& unfold(const Tensor& src, int dimension, int64_t size, int64_t step) = 0; |
| virtual Tensor& squeeze(const Tensor& src) = 0; |
| virtual Tensor& squeeze(const Tensor& src, int dimension) = 0; |
| virtual Tensor& unsqueeze(const Tensor& src, int dimension) = 0; |
| |
| virtual Tensor& copy(const Tensor& src) = 0; |
| virtual Tensor& gesv(const Tensor& ra, const Tensor& b, const Tensor& a) = 0; |
| virtual Tensor& trtrs(const Tensor& ra, const Tensor& b, const Tensor& a, |
| const char *uplo, const char *trans, const char *diag) = 0; |
| virtual Tensor& gels(const Tensor& ra, const Tensor& b, const Tensor& a) = 0; |
| virtual Tensor& syev(const Tensor& rv, const Tensor& a, |
| const char *jobz, const char *uplo) = 0; |
| virtual Tensor& geev(const Tensor& rv, const Tensor& a, const char *jobvr) = 0; |
| virtual Tensor& gesvd(const Tensor& rs, const Tensor& rv, |
| const Tensor& a, const char *jobu) = 0; |
| virtual Tensor& gesvd2(const Tensor& rs, const Tensor& rv, const Tensor& ra, |
| const Tensor& a, const char *jobu) = 0; |
| virtual Tensor& getri(const Tensor& a) = 0; |
| virtual Tensor& potrf(const Tensor& a, const char *uplo) = 0; |
| virtual Tensor& potrs(const Tensor& b, const Tensor& a, const char *uplo) = 0; |
| virtual Tensor& potri(const Tensor& a, const char *uplo) = 0; |
| virtual Tensor& qr(const Tensor& rr, const Tensor& a) = 0; |
| virtual Tensor& geqrf(const Tensor& rtau, const Tensor& a) = 0; |
| virtual Tensor& orgqr(const Tensor& a, const Tensor& tau) = 0; |
| virtual Tensor& ormqr(const Tensor& a, const Tensor& tau, const Tensor& c, |
| const char *side, const char *trans) = 0; |
| |
| virtual Tensor& cat(const std::vector<Tensor*>& src, int dimension) = 0; |
| virtual Tensor& gather(const Tensor& src, int dimension, const Tensor& index) = 0; |
| virtual Tensor& scatter(int dimension, const Tensor& index, const Tensor& src) = 0; |
| virtual Tensor& neg(const Tensor& src) = 0; |
| virtual Tensor& cadd(const Tensor& src1, const Tensor& src2) = 0; |
| virtual Tensor& cinv(const Tensor& src) = 0; |
| virtual Tensor& cmul(const Tensor& src1, const Tensor& src2) = 0; |
| virtual Tensor& cpow(const Tensor& src1, const Tensor& src2) = 0; |
| virtual Tensor& cdiv(const Tensor& src1, const Tensor& src2) = 0; |
| virtual Tensor& cfmod(const Tensor& src1, const Tensor& src2) = 0; |
| virtual Tensor& cremainder(const Tensor& src1, const Tensor& src2) = 0; |
| virtual Tensor& max(const Tensor& indices_, const Tensor& src, int dimension, int keepdim) = 0; |
| virtual Tensor& min(const Tensor& indices_, const Tensor& src, int dimension, int keepdim) = 0; |
| virtual Tensor& kthvalue(const Tensor& indices_, const Tensor& src, int64_t k, int dimension, int keepdim) = 0; |
| virtual Tensor& mode(const Tensor& indices_, const Tensor& src, int dimension, int keepdim) = 0; |
| virtual Tensor& median(const Tensor& indices_, const Tensor& src, int dimension, int keepdim) = 0; |
| virtual Tensor& sum(const Tensor& src, int dimension, int keepdim) = 0; |
| virtual Tensor& prod(const Tensor& src, int dimension, int keepdim) = 0; |
| virtual Tensor& cumsum(const Tensor& src, int dimension) = 0; |
| virtual Tensor& cumprod(const Tensor& src, int dimension) = 0; |
| virtual Tensor& sign(const Tensor& source) = 0; |
| virtual Tensor& cross(const Tensor& src1, const Tensor& src2, int dimension) = 0; |
| virtual Tensor& cmax(const Tensor& src1, const Tensor& src2) = 0; |
| virtual Tensor& cmin(const Tensor& src1, const Tensor& src2) = 0; |
| virtual Tensor& zero() = 0; |
| |
| virtual Tensor& maskedCopy(const Tensor& mask, const Tensor& src) = 0; |
| virtual Tensor& maskedSelect(const Tensor& mask, const Tensor& src) = 0; |
| // NOTE like in byte comparison operations, the order of arguments |
| // in nonzero is reversed compared to TH |
| virtual Tensor& nonzero(const Tensor& subscript) = 0; |
| virtual Tensor& indexSelect(const Tensor& src, int dim, const Tensor& index) = 0; |
| virtual Tensor& indexCopy(int dim, const Tensor& index, const Tensor& src) = 0; |
| virtual Tensor& indexAdd(int dim, const Tensor& index, const Tensor& src) = 0; |
| |
| virtual Tensor& diag(const Tensor& src, int k) = 0; |
| virtual Tensor& eye(int64_t n, int64_t m) = 0; |
| virtual Tensor& randperm(const Generator& _generator, int64_t n) = 0; |
| virtual Tensor& sort(const Tensor& ri, const Tensor& src, |
| int dimension, int desc) = 0; |
| virtual Tensor& topk(const Tensor& ri, const Tensor& src, |
| int64_t k, int dim, int dir, int sorted) = 0; |
| virtual Tensor& tril(const Tensor& src, int64_t k) = 0; |
| virtual Tensor& triu(const Tensor& src, int64_t k) = 0; |
| virtual Tensor& catArray(const std::vector<Tensor*>& inputs, int dimension) = 0; |
| virtual int equal(const Tensor& other) const = 0; |
| virtual Tensor& ltTensor(const Tensor& r, const Tensor& tb) = 0; |
| virtual Tensor& leTensor(const Tensor& r, const Tensor& tb) = 0; |
| virtual Tensor& gtTensor(const Tensor& r, const Tensor& tb) = 0; |
| virtual Tensor& geTensor(const Tensor& r, const Tensor& tb) = 0; |
| virtual Tensor& neTensor(const Tensor& r, const Tensor& tb) = 0; |
| virtual Tensor& eqTensor(const Tensor& r, const Tensor& tb) = 0; |
| virtual Tensor& ltTensorT(const Tensor& ta, const Tensor& tb) = 0; |
| virtual Tensor& leTensorT(const Tensor& ta, const Tensor& tb) = 0; |
| virtual Tensor& gtTensorT(const Tensor& ta, const Tensor& tb) = 0; |
| virtual Tensor& geTensorT(const Tensor& ta, const Tensor& tb) = 0; |
| virtual Tensor& neTensorT(const Tensor& ta, const Tensor& tb) = 0; |
| virtual Tensor& eqTensorT(const Tensor& ta, const Tensor& tb) = 0; |
| virtual Tensor& abs(const Tensor& src) = 0; |
| virtual Tensor& sigmoid(const Tensor& src) = 0; |
| virtual Tensor& log(const Tensor& src) = 0; |
| virtual Tensor& log1p(const Tensor& src) = 0; |
| virtual Tensor& exp(const Tensor& src) = 0; |
| virtual Tensor& cos(const Tensor& src) = 0; |
| virtual Tensor& acos(const Tensor& src) = 0; |
| virtual Tensor& cosh(const Tensor& src) = 0; |
| virtual Tensor& sin(const Tensor& src) = 0; |
| virtual Tensor& asin(const Tensor& src) = 0; |
| virtual Tensor& sinh(const Tensor& src) = 0; |
| |
| virtual Tensor& tan(const Tensor& src) = 0; |
| virtual Tensor& atan(const Tensor& src) = 0; |
| virtual Tensor& atan2(const Tensor& src1, const Tensor& src2) = 0; |
| virtual Tensor& tanh(const Tensor& src) = 0; |
| virtual Tensor& sqrt(const Tensor& src) = 0; |
| virtual Tensor& rsqrt(const Tensor& src) = 0; |
| virtual Tensor& ceil(const Tensor& src) = 0; |
| virtual Tensor& floor(const Tensor& src) = 0; |
| virtual Tensor& round(const Tensor& src) = 0; |
| virtual Tensor& trunc(const Tensor& src) = 0; |
| virtual Tensor& frac(const Tensor& src) = 0; |
| virtual Tensor& mean(const Tensor& src, int dimension, int keepdim) = 0; |
| virtual Tensor& std(const Tensor& src, int dimension, int biased, int keepdim) = 0; |
| virtual Tensor& var(const Tensor& src, int dimension, int biased, int keepdim) = 0; |
| virtual Tensor& rand(const Generator& _generator, THLongStorage *size) = 0; |
| virtual Tensor& randn(const Generator& _generator, THLongStorage *size) = 0; |
| |
| virtual int logicalall() = 0; |
| virtual int logicalany() = 0; |
| virtual Tensor& random(const Generator& _generator) = 0; |
| virtual Tensor& geometric(const Generator& _generator, double p) = 0; |
| virtual Tensor& bernoulli(const Generator& _generator, double p) = 0; |
| virtual Tensor& bernoulli_FloatTensor(const Generator& _generator, const Tensor& p) = 0; |
| virtual Tensor& bernoulli_DoubleTensor(const Generator& _generator, const Tensor& p) = 0; |
| virtual Tensor& uniform(const Generator& _generator, double a, double b) = 0; |
| virtual Tensor& normal(const Generator& _generator, double mean, double stdv) = 0; |
| virtual Tensor& exponential(const Generator& _generator, double lambda) = 0; |
| virtual Tensor& cauchy(const Generator& _generator, double median, double sigma) = 0; |
| virtual Tensor& logNormal(const Generator& _generator, double mean, double stdv) = 0; |
| |
| // Note: the order of *Tensor and *Prob_dist is reversed compared to |
| // the declarations in TH/generic/THTensorMath.h, so for instance |
| // the call: |
| // THRealTensor_multinomial(r, _generator, prob_dist, n_sample, with_replacement) |
| // is equivalent to `prob_dist->multinomial(r, _generator, n_sample, with_replacement)`. |
| // It is done this way so that the first argument can be casted onto a float tensor type. |
| virtual Tensor& multinomial(const Tensor& r, const Generator& _generator, |
| int n_sample, int with_replacement) = 0; |
| |
| virtual thpp::Type type() const = 0; |
| virtual bool isCuda() const = 0; |
| virtual bool isSparse() const = 0; |
| virtual int getDevice() const = 0; |
| virtual std::unique_ptr<Tensor> newTensor() const = 0; |
| }; |
| |
| template<typename real> |
| struct TensorScalarInterface : public Tensor { |
| using Tensor::cadd; |
| using scalar_type = real; |
| virtual TensorScalarInterface& fill(scalar_type value) = 0; |
| virtual TensorScalarInterface& maskedFill(const Tensor& mask, scalar_type value) = 0; |
| virtual TensorScalarInterface& indexFill(int dim, const Tensor& index, scalar_type value) = 0; |
| virtual TensorScalarInterface& scatterFill(int dimension, const Tensor& index, |
| scalar_type value) = 0; |
| virtual scalar_type dot(const Tensor& source) = 0; |
| virtual scalar_type minall() = 0; |
| virtual scalar_type maxall() = 0; |
| virtual scalar_type medianall() = 0; |
| virtual scalar_type sumall() = 0; |
| virtual scalar_type prodall() = 0; |
| virtual TensorScalarInterface& add(const Tensor& src, scalar_type value) = 0; |
| virtual TensorScalarInterface& sub(const Tensor& src, scalar_type value) = 0; |
| virtual TensorScalarInterface& mul(const Tensor& src, scalar_type value) = 0; |
| virtual TensorScalarInterface& div(const Tensor& src, scalar_type value) = 0; |
| virtual TensorScalarInterface& fmod(const Tensor& src, scalar_type value) = 0; |
| virtual TensorScalarInterface& remainder(const Tensor& src, scalar_type value) = 0; |
| virtual TensorScalarInterface& clamp(const Tensor& src, scalar_type min_value, |
| scalar_type max_value) = 0; |
| virtual TensorScalarInterface& cadd(const Tensor& src1, scalar_type value, |
| const Tensor& src2) = 0; |
| virtual TensorScalarInterface& csub(const Tensor& src1, scalar_type value, |
| const Tensor& src2) = 0; |
| virtual TensorScalarInterface& addcmul(const Tensor& src1, scalar_type value, |
| const Tensor& src2, |
| const Tensor& src3) = 0; |
| virtual TensorScalarInterface& addcdiv(const Tensor& src1, scalar_type value, |
| const Tensor& src2, |
| const Tensor& src3) = 0; |
| virtual TensorScalarInterface& addmv(scalar_type beta, const Tensor& src, |
| scalar_type alpha, const Tensor& mat, |
| const Tensor& vec) = 0; |
| virtual TensorScalarInterface& addmm(scalar_type beta, const Tensor& src, |
| scalar_type alpha, const Tensor& mat1, |
| const Tensor& mat2) = 0; |
| virtual TensorScalarInterface& addr(scalar_type beta, const Tensor& src, |
| scalar_type alpha, const Tensor& vec1, |
| const Tensor& vec2) = 0; |
| virtual TensorScalarInterface& addbmm(scalar_type beta, const Tensor& src, |
| scalar_type alpha, const Tensor& batch1, |
| const Tensor& batch2) = 0; |
| virtual TensorScalarInterface& baddbmm(scalar_type beta, const Tensor& src, |
| scalar_type alpha, const Tensor& batch1, |
| const Tensor& batch2) = 0; |
| virtual TensorScalarInterface& match(const Tensor& m1, const Tensor& m2, |
| scalar_type gain) = 0; |
| virtual scalar_type trace() = 0; |
| virtual TensorScalarInterface& cmaxValue(const Tensor& src, |
| scalar_type value) = 0; |
| virtual TensorScalarInterface& cminValue(const Tensor& src, |
| scalar_type value) = 0; |
| |
| virtual TensorScalarInterface& range(scalar_type xmin, scalar_type xmax, |
| scalar_type step) = 0; |
| virtual TensorScalarInterface& ltValue(const Tensor& t, scalar_type value) = 0; |
| virtual TensorScalarInterface& leValue(const Tensor& t, scalar_type value) = 0; |
| virtual TensorScalarInterface& gtValue(const Tensor& t, scalar_type value) = 0; |
| virtual TensorScalarInterface& geValue(const Tensor& t, scalar_type value) = 0; |
| virtual TensorScalarInterface& neValue(const Tensor& t, scalar_type value) = 0; |
| virtual TensorScalarInterface& eqValue(const Tensor& t, scalar_type value) = 0; |
| virtual TensorScalarInterface& ltValueT(const Tensor& t, scalar_type value) = 0; |
| virtual TensorScalarInterface& leValueT(const Tensor& t, scalar_type value) = 0; |
| virtual TensorScalarInterface& gtValueT(const Tensor& t, scalar_type value) = 0; |
| virtual TensorScalarInterface& geValueT(const Tensor& t, scalar_type value) = 0; |
| virtual TensorScalarInterface& neValueT(const Tensor& t, scalar_type value) = 0; |
| virtual TensorScalarInterface& eqValueT(const Tensor& t, scalar_type value) = 0; |
| |
| virtual TensorScalarInterface& pow(const Tensor& src, scalar_type value) = 0; |
| virtual TensorScalarInterface& tpow(scalar_type value, const Tensor& src) = 0; |
| virtual TensorScalarInterface& lerp(const Tensor& a, const Tensor& b, scalar_type weight) = 0; |
| virtual TensorScalarInterface& norm(const Tensor& src, scalar_type value, int dimension, int keepdim) = 0; |
| virtual TensorScalarInterface& renorm(const Tensor& src, scalar_type value, int dimension, scalar_type maxnorm) = 0; |
| virtual TensorScalarInterface& histc(const Tensor& src, int64_t nbins, scalar_type minvalue, scalar_type maxvalue) = 0; |
| virtual TensorScalarInterface& bhistc(const Tensor& src, int64_t nbins, scalar_type minvalue, scalar_type maxvalue) = 0; |
| |
| virtual scalar_type dist(const Tensor& src, scalar_type value) = 0; |
| virtual scalar_type meanall() = 0; |
| virtual scalar_type varall(int biased) = 0; |
| virtual scalar_type stdall(int biased) = 0; |
| virtual scalar_type normall(scalar_type value) = 0; |
| virtual TensorScalarInterface& linspace(scalar_type a, scalar_type b, int64_t n) = 0; |
| virtual TensorScalarInterface& logspace(scalar_type a, scalar_type b, int64_t n) = 0; |
| virtual TensorScalarInterface& pstrf(const Tensor& rpiv, const Tensor& a, |
| const char *uplo, scalar_type tol) = 0; |
| }; |
| |
| using FloatTensor = TensorScalarInterface<double>; |
| using IntTensor = TensorScalarInterface<int64_t>; |
| |
| } // namespace thpp |
