diff --git a/unit_tests/data/Test_Tensor.cpp b/unit_tests/data/Test_Tensor.cpp
index 1104cda98d2966be724ad235c328e0db93f9d8e2..f26901dc357080eebbaa2c6bea1b727134bd143d 100644
--- a/unit_tests/data/Test_Tensor.cpp
+++ b/unit_tests/data/Test_Tensor.cpp
@@ -10,23 +10,405 @@
********************************************************************************/
#include <array>
-#include <cstddef>
-#include <cstdint> //std::uint16_t
-#include <random>
+#include <cstddef> // std::size_t
+#include <cstdint> // std::uint8_t, std::uint16_t, std::int32_t
+#include <numeric> // std::accumulate, std::inner_product
+#include <functional> // std::multiplies
+#include <random> // std::random_device, std::mt19937,
+ // std::uniform_int_distribution, std::uniform_real_distribution
+#include <set>
+#include <string>
#include <vector>
#include <catch2/catch_test_macros.hpp>
+#include "aidge/backend/cpu/data/TensorImpl.hpp"
+#include "aidge/data/Data.hpp"
#include "aidge/data/Tensor.hpp"
-#include "aidge/utils/TensorUtils.hpp"
#include "aidge/utils/ArrayHelpers.hpp"
+#include "aidge/utils/TensorUtils.hpp"
+#include "aidge/utils/Types.h"
namespace Aidge {
-TEST_CASE("[backend/cpu/data] Tensor", "[Tensor]") {
- SECTION("Constructor") {
+TEST_CASE("[core/data] Tensor(Construction)", "[Tensor][Constructor]") {
+ SECTION("Default constructor") {
+ Tensor T_default{};
+ REQUIRE((
+ (T_default.dataType() == DataType::Float32) &&
+ (T_default.size() == 1) &&
+ (T_default.dims() == std::vector<DimSize_t>({})) &&
+ (T_default.strides() == std::vector<DimSize_t>({1})) &&
+ (T_default.getImpl() == nullptr) &&
+ (T_default.grad() == nullptr) &&
+ (T_default.isContiguous() == true)
+ ));
+ }
+ SECTION("scalar constructor") {
+ Tensor T;
+ REQUIRE_NOTHROW(T = Tensor(std::int32_t(20)));
+ REQUIRE((
+ (T.dataType() == DataType::Int32) &&
+ (T.size() == 1) &&
+ (T.dims() == std::vector<DimSize_t>({})) &&
+ (T.strides() == std::vector<DimSize_t>({1})) &&
+ (T.getImpl() != nullptr) &&
+ (T.grad() == nullptr) &&
+ (T.isContiguous() == true)
+ ));
+ }
+ SECTION("dim constructor") {
+ const std::vector<DimSize_t> Tdims = {1,2,3,4,5,6,7};
+ Tensor T;
+ REQUIRE_NOTHROW(T = Tensor(Tdims));
+ REQUIRE((
+ (T.dataType() == DataType::Float32) &&
+ (T.size() == std::accumulate(Tdims.cbegin(), Tdims.cend(), DimSize_t(1), std::multiplies<DimSize_t>())) &&
+ (T.dims() == Tdims) &&
+ (T.strides() == std::vector<DimSize_t>({5040,2520,840,210,42,7,1})) &&
+ (T.getImpl() == nullptr) &&
+ (T.grad() == nullptr) &&
+ (T.isContiguous() == true)
+ ));
+ }
+ SECTION("TensorUtils, constructor from const arrays") {
+ Tensor T;
+ // Construction from different types and sizes
+
+ // Set an already constructed Tensor
+ REQUIRE_NOTHROW(T = Array1D<int, 2>{{1, 2}});
+ REQUIRE((
+ (T.dataType() == DataType::Int32) &&
+ (T.size() == 2) &&
+ (T.dims() == std::vector<DimSize_t>({2})) &&
+ (T.strides() == std::vector<DimSize_t>({1})) &&
+ (T.getImpl() != nullptr) &&
+ (T.grad() == nullptr) &&
+ (T.isContiguous() == true)
+ ));
+
+ // Change dims
+ REQUIRE_NOTHROW(T = Array2D<int, 2, 2>{{{1, 2}, {3, 4}}});
+ // Change data types
+ REQUIRE_NOTHROW(T = Array3D<std::uint8_t, 2, 2, 2>{{{{1, 2}, {3, 4}}, {{5, 6}, {7, 8}}}});
+ REQUIRE((
+ (T.dataType() == DataType::UInt8) &&
+ (T.size() == 8) &&
+ (T.dims() == std::vector<DimSize_t>({2,2,2})) &&
+ (T.strides() == std::vector<DimSize_t>({4,2,1})) &&
+ (T.getImpl() != nullptr) &&
+ (T.grad() == nullptr) &&
+ (T.isContiguous() == true)
+ ));
+ REQUIRE_NOTHROW(T = Array3D<int, 2, 2, 2>{{{{1, 2}, {3, 4}}, {{5, 6}, {7, 8}}}});
+ REQUIRE_NOTHROW(T = Array3D<float, 2, 2, 2>{{{{1.0f, 2.0f}, {3.0f, 4.0f}}, {{5.0f, 6.0f}, {7.0f, 8.0f}}}});
+ REQUIRE_NOTHROW(T = Array3D<double, 2, 2, 2>{{{{1., 2.}, {3., 4.}}, {{5., 6.}, {7., 8.}}}});
+
+ // Change dims
+ REQUIRE_NOTHROW(T = Array4D<int, 2, 2, 2, 2>{{{{{1, 2}, {3, 4}}, {{5, 6}, {7, 8}}},
+ {{{9,10}, {11,12}}, {{13,14},{15,16}}}}});
+ REQUIRE((
+ (T.dataType() == DataType::Int32) &&
+ (T.size() == 16) &&
+ (T.dims() == std::vector<DimSize_t>({2,2,2,2})) &&
+ (T.strides() == std::vector<DimSize_t>({8,4,2,1})) &&
+ (T.getImpl() != nullptr) &&
+ (T.grad() == nullptr) &&
+ (T.isContiguous() == true)
+ ));
+ }
+ SECTION("copy constructor / copy assignment operator") {
+
+ }
+ SECTION("move constructor / move assignment operator") {
+
+ }
+ SECTION("prototype") {
+ constexpr std::uint16_t NBTRIALS = 10;
+
+ // Create random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<std::size_t> dimsDist(1, 10);
+ std::uniform_int_distribution<std::size_t> nbDimsDist(1, 5);
+ std::uniform_real_distribution<float> valueDist(0.001f, 1.0f);
+
+ for (std::size_t trial = 0; trial < NBTRIALS; ++trial) {
+ std::vector<std::size_t> Tdims;
+ const std::size_t Tsize = nbDimsDist(gen);
+ for (std::size_t i = 0; i < Tsize; ++i) {
+ Tdims.push_back(dimsDist(gen));
+ }
+ Tensor T(Tdims);
+
+ // file the tensor
+ float* array0 = new float[T.size()];
+ for (std::size_t i = 0; i < T.size(); ++i) {
+ array0[i] = valueDist(gen);
+ }
+ T.setBackend("cpu");
+ T.getImpl() -> setRawPtr(array0, T.size());
+
+ Tensor Tclone;
+ REQUIRE_NOTHROW(Tclone = T.clone());
+ REQUIRE((
+ (T.dataType() == Tclone.dataType()) &&
+ (T.size() == Tclone.size()) &&
+ (T.dims() == Tclone.dims()) &&
+ (T.strides() == Tclone.strides()) &&
+ (T.getImpl() != Tclone.getImpl()) &&
+ (Tclone.grad() == nullptr) &&
+ (Tclone.isContiguous() == true)
+ ));
+ REQUIRE(Tclone == T);
+ }
+ }
+}
+
+TEST_CASE("[core/data] Tensor(getter/setter)", "[Tensor][Getter][Setter]") {
+ constexpr std::uint16_t NBTRIALS = 10;
+
+ // Create random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<std::size_t> dimsDist(1, 10);
+ std::uniform_int_distribution<std::size_t> nbDimsDist(1, 5);
+ std::uniform_real_distribution<float> valueDist(0.001f, 1.0f);
+
+ for (std::size_t trial = 0; trial < NBTRIALS; ++trial) {
+ std::vector<std::size_t> Tdims;
+ const std::size_t Tsize = nbDimsDist(gen);
+ for (std::size_t i = 0; i < Tsize; ++i) {
+ Tdims.push_back(dimsDist(gen));
+ }
+
+ // create Tensor
+ Tensor T(Tdims);
+ // compute stride
+ std::vector<std::size_t> Tstrides(Tdims.size(), 1);
+ std::size_t i = Tdims.size() - 1;
+ while (i-- > 0) {
+ Tstrides[i] = Tstrides[i+1]*Tdims[i+1];
+ }
+
+ /////////////////
+ // dimensions
+ // nbDims(), dims(), size()
+ REQUIRE(T.nbDims() == Tdims.size());
+
+ REQUIRE(T.dims() == Tdims);
+
+ std::size_t trueSize = std::accumulate(Tdims.cbegin(), Tdims.cend(), 1, std::multiplies<std::size_t>());
+ REQUIRE(T.size() == trueSize);
+
+ /////////////////
+ // implementation
+ // getImpl(), setImpl(), hasImpl()
+ REQUIRE(T.hasImpl() == false);
+ std::shared_ptr<TensorImpl_cpu<float>> tensorImpl = std::make_shared<TensorImpl_cpu<float>>(0, Tdims);
+
+ T.setImpl(tensorImpl);
+ REQUIRE(T.getImpl() == tensorImpl);
+ REQUIRE(T.hasImpl() == true);
+
+ // isContiguous(), stride(),
+ REQUIRE(T.isContiguous());
+ REQUIRE(T.strides() == Tstrides);
+
+ // file the tensor
+ float* array0 = new float[T.size()];
+ for (std::size_t i = 0; i < T.size(); ++i) {
+ array0[i] = valueDist(gen);
+ }
+ tensorImpl -> setRawPtr(array0, T.size());
+
+ // getCoord(), getIdx(), getStorageIdx()
+ std::vector<DimSize_t> Tdims_copy = Tdims;
+ for (auto& val : Tdims_copy) {
+ val = std::min(DimSize_t(2), std::max(DimSize_t(0), val - 1));
+ }
+ DimSize_t true_flatid = std::inner_product(Tdims_copy.cbegin(), Tdims_copy.cend(), Tstrides.cbegin(), DimSize_t(0));
+
+ REQUIRE(T.getCoord(true_flatid) == Tdims_copy);
+ REQUIRE(T.getIdx(Tdims_copy) == true_flatid);
+ REQUIRE(T.getStorageIdx(Tdims_copy) == true_flatid); // Tensor is not a view
+
+ // set(vector), set(size_t), get(vector), get(size_t), getImplOffset()
+ REQUIRE_NOTHROW(T.set<float>(Tdims_copy, 50.0f));
+ REQUIRE(T.get<float>(Tdims_copy) == 50.0f);
+
+ REQUIRE_NOTHROW(T.set<float>(true_flatid, 40.0f));
+ REQUIRE(T.get<float>(true_flatid) == 40.0f);
+ REQUIRE(T.getImplOffset() == 0);
+
+ delete[] array0;
+
+ //////////////
+ // backend
+ // getAvailableBackends()
+ REQUIRE(Tensor::getAvailableBackends() == std::set<std::string>({"cpu"}));
+
+ // setBackend()
+ REQUIRE_NOTHROW(T.setBackend("cpu", 0));
+
+ // setDataType(), dataType()
+ REQUIRE_NOTHROW(T.setDataType(DataType::Int16));
+ REQUIRE(T.dataType() == DataType::Int16);
+ }
+}
+TEST_CASE("[core/data] Tensor(other)", "[Tensor][extract][zeros][print]") {
+ // extract, makeContiguous
+ // empty
+ constexpr std::uint16_t NBTRIALS = 10;
+
+ // Create random number generators
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ std::uniform_int_distribution<std::size_t> dimsDist(1, 10);
+ std::uniform_int_distribution<std::size_t> nbDimsDist(1, 5);
+ std::uniform_real_distribution<float> valueDist(0.001f, 1.0f);
+ // zeros, resize
+ SECTION("zeros") {
+ Tensor T;
+ for (std::size_t trial = 0; trial < NBTRIALS; ++trial) {
+ std::vector<std::size_t> Tdims;
+ const std::size_t Tsize = nbDimsDist(gen);
+ for (std::size_t i = 0; i < Tsize; ++i) {
+ Tdims.push_back(dimsDist(gen));
+ }
+ T.resize(Tdims);
+
+ // file the tensor
+ float* array0 = new float[T.size()];
+ for (std::size_t i = 0; i < T.size(); ++i) {
+ array0[i] = valueDist(gen);
+ }
+ T.setBackend("cpu");
+ T.getImpl() -> setRawPtr(array0, T.size());
+ float* res = static_cast<float*>(T.getImpl()->hostPtr());
+ for (std::size_t i = 0; i < T.size(); ++i) {
+ REQUIRE(res[i] == array0[i]);
+ }
+
+ T.zeros();
+ res = static_cast<float*>(T.getImpl()->hostPtr());
+ for (std::size_t i = 0; i < T.size(); ++i) {
+ REQUIRE(res[i] == 0.0f);
+ }
+ delete[] array0;
+ }
+ }
+
+ SECTION("Tensor extract") {
+ bool equal;
+
+ for (std::uint16_t trial = 0; trial < NBTRIALS; ++trial) {
+ // create Tensor
+ const std::size_t nb_dims = 3;
+ const std::size_t dim0 = dimsDist(gen) + 1; // dim0 >= 2
+ const std::size_t dim1 = dimsDist(gen) + 1;
+ const std::size_t dim2 = dimsDist(gen) + 1;
+ std::vector<std::size_t> dims = {dim0, dim1, dim2};
+ int array0[dim0*dim1*dim2];
+ for (std::size_t i = 0; i < dim0; ++i) {
+ for (std::size_t j = 0; j < dim1; ++j) {
+ for (std::size_t k = 0; k < dim2; ++k) {
+ array0[((i * dim1) + j)*dim2 + k] = valueDist(gen);
+ }
+ }
+ }
+ Tensor x{dims};
+ x.setDataType(DataType::Int32);
+ x.setBackend("cpu");
+ Tensor y;
+ Tensor y0;
+ Tensor y1;
+ Tensor y2;
+ Tensor y3;
+ x.getImpl()->setRawPtr(&array0, dim0*dim1*dim2);
+ REQUIRE(x.isContiguous());
+
+ ////////////////
+ // extract contiguous Tensor slice given start coordinates
+ // the whole Tensor
+ REQUIRE_NOTHROW(y0 = x.extract({}));
+ REQUIRE(y0 == x);
+ int* y0_res = static_cast<int*>(y0.getImpl()->hostPtr());
+ equal = true;
+ for (std::size_t i = 0; i < dim0*dim1*dim2; ++i) {
+ equal &= (y0_res[i] == array0[i]);
+ }
+ REQUIRE(equal);
+ REQUIRE(y0.getImpl() == x.getImpl());
+ REQUIRE(y0.isContiguous());
+
+ // Tensor - 1-D
+ REQUIRE_NOTHROW(y1 = x.extract({dim0 - 2}));
+ int* y1_res = static_cast<int*>(y1.getImpl()->hostPtr());
+ equal = true;
+ for (std::size_t i = 0; i < dim1*dim2; ++i) {
+ equal &= (y1_res[i] == array0[(dim0-2)*dim1*dim2 + i]);
+ }
+ REQUIRE(equal);
+ REQUIRE(y1.getImpl() == x.getImpl());
+ REQUIRE(y1.isContiguous());
+
+ // Tensor - 2-D
+ REQUIRE_NOTHROW(y2 = x.extract({dim0 - 2, dim1 - 2}));
+ int* y2_res = static_cast<int*>(y2.getImpl()->hostPtr());
+ equal = true;
+ for (std::size_t i = 0; i < dim2; ++i) {
+ equal &= (y2_res[i] == array0[(((dim0 - 2) * dim1) + (dim1 - 2))*dim2 + i]);
+ }
+ REQUIRE(equal);
+ REQUIRE(y2.getImpl() == x.getImpl());
+ REQUIRE(y2.isContiguous());
+
+ // Tensor - 3-D => scalar
+ REQUIRE_NOTHROW(y3 = x.extract({dim0 - 2, dim1 - 2, dim2 - 2}));
+ int* y3_res = static_cast<int*>(y3.getImpl()->hostPtr());
+ REQUIRE(y3_res[0] == array0[(((dim0 - 2) * dim1) + (dim1 - 2))*dim2 + dim2 - 2]);
+ REQUIRE(y3.getImpl() == x.getImpl());
+ REQUIRE(y3.isContiguous());
+
+ // throw an error
+ REQUIRE_THROWS(y = x.extract({0, dim1, 0}));
+
+ /////////////////
+ // extract Tensor slice given start coordinates and dimension
+ REQUIRE_NOTHROW(y = x.extract({0, 0, 1}, {dim0-1, 1, dim2-1}));
+ REQUIRE(y.getImpl() == x.getImpl()); // shared implem
+ REQUIRE(!y.isContiguous());
+
+ Tensor yClone = y.clone(); // when copying data, they are contiguous in memory
+ REQUIRE(yClone.isContiguous());
+ // int yTruth[2][1][1] =
+ REQUIRE(approxEq<int>(yClone, y, 0.0f, 0.0f));
+ }
+ }
+ // print, toString,
+ SECTION("Pretty printing for debug") {
+ Tensor x{};
+ // Empty Tensor
+ REQUIRE_THROWS(x.print());
+ // scalar
+ x = Tensor(42);
+ REQUIRE_NOTHROW(x.print());
+ // 1-D Tensors
+ x = Array1D<int, 1>{{1}};
+ REQUIRE_NOTHROW(x.print());
+ x = Array1D<int, 6>{{1,2,3,4,5,6}};
+ REQUIRE_NOTHROW(x.print());
+ // 2-D Tensors
+ x = Array2D<int, 3, 2>{{{1, 2}, {3, 4}, {5, 6}}};
+ REQUIRE_NOTHROW(x.print());
+ // +2-D Tensors
+ x = Array3D<int, 2, 2, 2>{{{{1, 2}, {3, 4}}, {{5, 6}, {7, 8}}}};
+ REQUIRE_NOTHROW(x.print());
+ x = Array4D<int, 2, 2, 2, 2>{{{{{1, 2}, {3, 4}}, {{5, 6}, {7, 8}}},{{{11, 12}, {13, 14}}, {{15, 16}, {17, 18}}}}};
+ REQUIRE_NOTHROW(x.print());
}
}
-} // namespace Aidge
\ No newline at end of file
+} // namespace Aidge