diff --git a/include/aidge/backend/cpu/operator/TransposeImpl.hpp b/include/aidge/backend/cpu/operator/TransposeImpl.hpp
index 3bcbeda6b6263fcbe7e33cd907b8f13bd62b6471..712e672752648f5ff8a3c073f6c81bbe7cc85d9d 100644
--- a/include/aidge/backend/cpu/operator/TransposeImpl.hpp
+++ b/include/aidge/backend/cpu/operator/TransposeImpl.hpp
@@ -24,22 +24,34 @@ namespace Aidge {
// compute kernel registry for forward and backward
class TransposeImpl2DForward_cpu
- : public Registrable<TransposeImpl2DForward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<2>::Attrs& attrs, const std::vector<DimSize_t>, const std::vector<DimSize_t>,const void*, void*)> {
+ : public Registrable<TransposeImpl2DForward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<2>::Attrs& attrs, const std::vector<DimSize_t>&, const std::vector<DimSize_t>&, const void*, void*)> {
};
class TransposeImpl3DForward_cpu
- : public Registrable<TransposeImpl3DForward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<3>::Attrs& attrs, const std::vector<DimSize_t>, const std::vector<DimSize_t>,const void*, void*)> {
+ : public Registrable<TransposeImpl3DForward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<3>::Attrs& attrs, const std::vector<DimSize_t>&, const std::vector<DimSize_t>&, const void*, void*)> {
};
class TransposeImpl4DForward_cpu
- : public Registrable<TransposeImpl4DForward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<4>::Attrs& attrs, const std::vector<DimSize_t>, const std::vector<DimSize_t>,const void*, void*)> {
+ : public Registrable<TransposeImpl4DForward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<4>::Attrs& attrs, const std::vector<DimSize_t>&, const std::vector<DimSize_t>&, const void*, void*)> {
+};
+class TransposeImpl5DForward_cpu
+ : public Registrable<TransposeImpl5DForward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<5>::Attrs& attrs, const std::vector<DimSize_t>&, const std::vector<DimSize_t>&, const void*, void*)> {
+};
+class TransposeImpl6DForward_cpu
+ : public Registrable<TransposeImpl6DForward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<6>::Attrs& attrs, const std::vector<DimSize_t>&, const std::vector<DimSize_t>&, const void*, void*)> {
};
class TransposeImpl2DBackward_cpu
- : public Registrable<TransposeImpl2DBackward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<2>::Attrs& attrs, const std::vector<DimSize_t>, const std::vector<DimSize_t>,const void*, void*)> {
+ : public Registrable<TransposeImpl2DBackward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<2>::Attrs& attrs, const std::vector<DimSize_t>&, const std::vector<DimSize_t>&, const void*, void*)> {
};
class TransposeImpl3DBackward_cpu
- : public Registrable<TransposeImpl3DBackward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<3>::Attrs& attrs, const std::vector<DimSize_t>, const std::vector<DimSize_t>,const void*, void*)> {
+ : public Registrable<TransposeImpl3DBackward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<3>::Attrs& attrs, const std::vector<DimSize_t>&, const std::vector<DimSize_t>&, const void*, void*)> {
};
class TransposeImpl4DBackward_cpu
- : public Registrable<TransposeImpl4DBackward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<4>::Attrs& attrs, const std::vector<DimSize_t>, const std::vector<DimSize_t>,const void*, void*)> {
+ : public Registrable<TransposeImpl4DBackward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<4>::Attrs& attrs, const std::vector<DimSize_t>&, const std::vector<DimSize_t>&, const void*, void*)> {
+};
+class TransposeImpl5DBackward_cpu
+ : public Registrable<TransposeImpl5DBackward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<5>::Attrs& attrs, const std::vector<DimSize_t>&, const std::vector<DimSize_t>&, const void*, void*)> {
+};
+class TransposeImpl6DBackward_cpu
+ : public Registrable<TransposeImpl6DBackward_cpu, std::tuple<DataType, DataType>, void( const typename Transpose_Op<6>::Attrs& attrs, const std::vector<DimSize_t>&, const std::vector<DimSize_t>&, const void*, void*)> {
};
@@ -76,11 +88,35 @@ public:
NbElts_t getNbRequiredProtected(const IOIndex_t inputIdx) const override final;
void forward() override;
};
+class TransposeImpl5D_cpu : public OperatorImpl {
+public:
+ TransposeImpl5D_cpu(const Transpose_Op<5>& op) : OperatorImpl(op) {}
+
+ static std::unique_ptr<TransposeImpl5D_cpu> create(const Transpose_Op<5>& op) {
+ return std::make_unique<TransposeImpl5D_cpu>(op);
+ }
+
+ NbElts_t getNbRequiredProtected(const IOIndex_t inputIdx) const override final;
+ void forward() override;
+};
+class TransposeImpl6D_cpu : public OperatorImpl {
+public:
+ TransposeImpl6D_cpu(const Transpose_Op<6>& op) : OperatorImpl(op) {}
+
+ static std::unique_ptr<TransposeImpl6D_cpu> create(const Transpose_Op<6>& op) {
+ return std::make_unique<TransposeImpl6D_cpu>(op);
+ }
+
+ NbElts_t getNbRequiredProtected(const IOIndex_t inputIdx) const override final;
+ void forward() override;
+};
namespace {
static Registrar<Transpose_Op<2>> registrarTransposeImpl2D_cpu("cpu", Aidge::TransposeImpl2D_cpu::create);
static Registrar<Transpose_Op<3>> registrarTransposeImpl3D_cpu("cpu", Aidge::TransposeImpl3D_cpu::create);
static Registrar<Transpose_Op<4>> registrarTransposeImpl4D_cpu("cpu", Aidge::TransposeImpl4D_cpu::create);
+static Registrar<Transpose_Op<5>> registrarTransposeImpl5D_cpu("cpu", Aidge::TransposeImpl5D_cpu::create);
+static Registrar<Transpose_Op<6>> registrarTransposeImpl6D_cpu("cpu", Aidge::TransposeImpl6D_cpu::create);
}
} // namespace Aidge
diff --git a/include/aidge/backend/cpu/operator/TransposeImpl_forward_kernels.hpp b/include/aidge/backend/cpu/operator/TransposeImpl_forward_kernels.hpp
index 79901424189b9327682376909bee087ae9eef47a..307b6d99e97dd3a4017ef4b45c109dec690a5e2a 100644
--- a/include/aidge/backend/cpu/operator/TransposeImpl_forward_kernels.hpp
+++ b/include/aidge/backend/cpu/operator/TransposeImpl_forward_kernels.hpp
@@ -22,7 +22,7 @@
namespace Aidge {
template <class I, class O, DimSize_t DIM>
-void TransposeImpl_cpu_forward_kernel( const typename Transpose_Op<DIM>::Attrs& attrs, const std::vector<DimSize_t> inputDims, const std::vector<DimSize_t> outputDims, const void* input_, void* output_)
+void TransposeImpl_cpu_forward_kernel( const typename Transpose_Op<DIM>::Attrs& attrs, const std::vector<DimSize_t>& inputDims, const std::vector<DimSize_t>& outputDims, const void* input_, void* output_)
{
O* output = static_cast<O*>(output_);
const I* input = static_cast<const I*>(input_);
@@ -87,6 +87,20 @@ static Registrar<TransposeImpl4DForward_cpu> registrarTransposeImpl4DForward_cpu
{DataType::Int32, DataType::Int32}, Aidge::TransposeImpl_cpu_forward_kernel<int, int, 4>);
static Registrar<TransposeImpl4DForward_cpu> registrarTransposeImpl4DForward_cpu_Float64(
{DataType::Float64, DataType::Float64}, Aidge::TransposeImpl_cpu_forward_kernel<double, double, 4>);
+// DIM = 5
+static Registrar<TransposeImpl5DForward_cpu> registrarTransposeImpl5DForward_cpu_Float32(
+ {DataType::Float32, DataType::Float32}, Aidge::TransposeImpl_cpu_forward_kernel<float, float, 5>);
+static Registrar<TransposeImpl5DForward_cpu> registrarTransposeImpl5DForward_cpu_Int32(
+ {DataType::Int32, DataType::Int32}, Aidge::TransposeImpl_cpu_forward_kernel<int, int, 5>);
+static Registrar<TransposeImpl5DForward_cpu> registrarTransposeImpl5DForward_cpu_Float64(
+ {DataType::Float64, DataType::Float64}, Aidge::TransposeImpl_cpu_forward_kernel<double, double, 5>);
+// DIM = 6
+static Registrar<TransposeImpl6DForward_cpu> registrarTransposeImpl6DForward_cpu_Float32(
+ {DataType::Float32, DataType::Float32}, Aidge::TransposeImpl_cpu_forward_kernel<float, float, 6>);
+static Registrar<TransposeImpl6DForward_cpu> registrarTransposeImpl6DForward_cpu_Int32(
+ {DataType::Int32, DataType::Int32}, Aidge::TransposeImpl_cpu_forward_kernel<int, int, 6>);
+static Registrar<TransposeImpl6DForward_cpu> registrarTransposeImpl6DForward_cpu_Float64(
+ {DataType::Float64, DataType::Float64}, Aidge::TransposeImpl_cpu_forward_kernel<double, double, 6>);
} // namespace
} // namespace Aidge
diff --git a/src/operator/TransposeImpl.cpp b/src/operator/TransposeImpl.cpp
index 7485351c67e5e79219b6d4ac4ca2d58606eeebe6..c3e8aaec12eef69675d35925dbca3c49fdca4065 100644
--- a/src/operator/TransposeImpl.cpp
+++ b/src/operator/TransposeImpl.cpp
@@ -25,17 +25,22 @@ Aidge::NbElts_t Aidge::TransposeImpl2D_cpu::getNbRequiredProtected(IOIndex_t /*i
// this implementation can be in-place
return 0;
}
-
Aidge::NbElts_t Aidge::TransposeImpl3D_cpu::getNbRequiredProtected(IOIndex_t /*inputIdx*/) const {
// this implementation can be in-place
return 0;
}
-
-
Aidge::NbElts_t Aidge::TransposeImpl4D_cpu::getNbRequiredProtected(IOIndex_t /*inputIdx*/) const {
// this implementation can be in-place
return 0;
}
+Aidge::NbElts_t Aidge::TransposeImpl5D_cpu::getNbRequiredProtected(IOIndex_t /*inputIdx*/) const {
+ // this implementation can be in-place
+ return 0;
+}
+Aidge::NbElts_t Aidge::TransposeImpl6D_cpu::getNbRequiredProtected(IOIndex_t /*inputIdx*/) const {
+ // this implementation can be in-place
+ return 0;
+}
void Aidge::TransposeImpl2D_cpu::forward() {
assert(mOp.getInput(0) && "missing input #0");
@@ -91,3 +96,33 @@ void Aidge::TransposeImpl4D_cpu::forward() {
mOp.getInput(0)->getImpl()->rawPtr(),
mOp.getOutput(0)->getImpl()->rawPtr());
}
+void Aidge::TransposeImpl5D_cpu::forward() {
+ assert(mOp.getInput(0) && "missing input #0");
+ assert(mOp.getInput(0)->nbDims() == 5 && "input #0 must have the same size as axes attributes (5)");
+
+ // Find the correct kernel type
+ auto kernelFunc =
+ Registrar<TransposeImpl5DForward_cpu>::create({mOp.getInput(0)->dataType(), mOp.getOutput(0)->dataType()});
+
+ // Call kernel
+ kernelFunc(dynamic_cast<const Transpose_Op<5>&>(mOp).getStaticAttributes(),
+ mOp.getInput(0)->dims(),
+ mOp.getOutput(0)->dims(),
+ mOp.getInput(0)->getImpl()->rawPtr(),
+ mOp.getOutput(0)->getImpl()->rawPtr());
+}
+void Aidge::TransposeImpl6D_cpu::forward() {
+ assert(mOp.getInput(0) && "missing input #0");
+ assert(mOp.getInput(0)->nbDims() == 6 && "input #0 must have the same size as axes attributes (6)");
+
+ // Find the correct kernel type
+ auto kernelFunc =
+ Registrar<TransposeImpl6DForward_cpu>::create({mOp.getInput(0)->dataType(), mOp.getOutput(0)->dataType()});
+
+ // Call kernel
+ kernelFunc(dynamic_cast<const Transpose_Op<6>&>(mOp).getStaticAttributes(),
+ mOp.getInput(0)->dims(),
+ mOp.getOutput(0)->dims(),
+ mOp.getInput(0)->getImpl()->rawPtr(),
+ mOp.getOutput(0)->getImpl()->rawPtr());
+}
\ No newline at end of file
diff --git a/unit_tests/operator/Test_TransposeImpl.cpp b/unit_tests/operator/Test_TransposeImpl.cpp
index 327d4f8db5570cd9258259eb1844b4b3fc74c607..5ec38e52a84d0c7d2693f52e1cc0d3851cb09946 100644
--- a/unit_tests/operator/Test_TransposeImpl.cpp
+++ b/unit_tests/operator/Test_TransposeImpl.cpp
@@ -17,89 +17,109 @@
#include "aidge/backend/cpu.hpp"
-#include <iostream>
using namespace Aidge;
TEST_CASE("[cpu/operator] Transpose(forward)") {
- std::shared_ptr<Tensor> input = std::make_shared<Tensor>(Array3D<float,2,3,4> {
- {
- {{0.42507452, 0.11244237, 0.43243718, 0.62354952},
- {0.90250170, 0.48719984, 0.45781207, 0.92536664},
- {0.06348717, 0.91678733, 0.64452291, 0.00484818}},
+ SECTION("3D Tensor") {
+ std::shared_ptr<Tensor> input = std::make_shared<Tensor>(Array3D<float,2,3,4> {
+ {
+ {{0.42507452, 0.11244237, 0.43243718, 0.62354952},
+ {0.90250170, 0.48719984, 0.45781207, 0.92536664},
+ {0.06348717, 0.91678733, 0.64452291, 0.00484818}},
- {{0.66873497, 0.99508536, 0.55714869, 0.84887981},
- {0.41666120, 0.92365038, 0.80034822, 0.38721532},
- {0.52037925, 0.53937608, 0.66380072, 0.36330253}}
- }
- });
- std::shared_ptr<Tensor> output = std::make_shared<Tensor>(Array3D<float,2,4,3> {
- {
- {{0.42507452, 0.90250170, 0.06348717},
- {0.11244237, 0.48719984, 0.91678733},
- {0.43243718, 0.45781207, 0.64452291},
- {0.62354952, 0.92536664, 0.00484818}},
+ {{0.66873497, 0.99508536, 0.55714869, 0.84887981},
+ {0.41666120, 0.92365038, 0.80034822, 0.38721532},
+ {0.52037925, 0.53937608, 0.66380072, 0.36330253}}
+ }
+ });
+ std::shared_ptr<Tensor> output = std::make_shared<Tensor>(Array3D<float,2,4,3> {
+ {
+ {{0.42507452, 0.90250170, 0.06348717},
+ {0.11244237, 0.48719984, 0.91678733},
+ {0.43243718, 0.45781207, 0.64452291},
+ {0.62354952, 0.92536664, 0.00484818}},
- {{0.66873497, 0.41666120, 0.52037925},
- {0.99508536, 0.92365038, 0.53937608},
- {0.55714869, 0.80034822, 0.66380072},
- {0.84887981, 0.38721532, 0.36330253}}
- }
- });
- std::shared_ptr<Node> myTranspose = Transpose<3>(std::array<DimSize_t,3>{{0,2,1}});
- myTranspose->getOperator()->setDatatype(DataType::Float32);
- myTranspose->getOperator()->setBackend("cpu");
- myTranspose->getOperator()->associateInput(0,input);
- myTranspose->getOperator()->computeOutputDims();
- myTranspose->forward();
+ {{0.66873497, 0.41666120, 0.52037925},
+ {0.99508536, 0.92365038, 0.53937608},
+ {0.55714869, 0.80034822, 0.66380072},
+ {0.84887981, 0.38721532, 0.36330253}}
+ }
+ });
+ std::shared_ptr<Node> myTranspose = Transpose<3>(std::array<DimSize_t,3>{{0,2,1}});
+ myTranspose->getOperator()->setDatatype(DataType::Float32);
+ myTranspose->getOperator()->setBackend("cpu");
+ myTranspose->getOperator()->associateInput(0,input);
+ myTranspose->getOperator()->computeOutputDims();
+ myTranspose->forward();
- // float* resPtr = static_cast<float*>(myTranspose->getOperator()->getOutput(0)->getImpl()->rawPtr());
- // float* expectedPtr = static_cast<float*>(output->getImpl()->rawPtr());
- // for (std::size_t i = 0; i< 24; ++i) {
- // std::cout << "Res " << resPtr[i] << " , expected : " << expectedPtr[i] << std::endl;
- // REQUIRE(std::abs(resPtr[i]-expectedPtr[i]) < 0.00001);
- // }
- REQUIRE(*(myTranspose->getOperator()->getOutput(0)) == *output);
+ REQUIRE(*(myTranspose->getOperator()->getOutput(0)) == *output);
+ }
+ SECTION("4D Tensor") {
+ std::shared_ptr<Tensor> input = std::make_shared<Tensor>(Array4D<int,2,3,1,4> {
+ {
+ {
+ {
+ {1, 2, 3, 4}
+ },
+ {
+ {5, 6, 7, 8}
+ },
+ {
+ {9, 10, 11, 12}
+ }
+ },
+ {
+ {
+ {13, 14, 15, 16}
+ },
+ {
+ {17, 18, 19, 20}
+ },
+ {
+ {21, 22, 23, 24}
+ }
+ }
+ }
+ });
+ std::shared_ptr<Tensor> output = std::make_shared<Tensor>(Array4D<int,2,4,1,3> {
+ {
+ {
+ {
+ {1, 5, 9}
+ },
+ {
+ {2, 6, 10}
+ },
+ {
+ {3, 7, 11}
+ },
+ {
+ {4, 8, 12}
+ }
+ },
+ {
+ {
+ {13, 17, 21}
+ },
+ {
+ {14, 18, 22}
+ },
+ {
+ {15, 19, 23}
+ },
+ {
+ {16, 20, 24}
+ }
+ }
+ }
+ });
+ std::shared_ptr<Node> myTranspose = Transpose<4>(std::array<DimSize_t,4>{{0,3,2,1}});
+ myTranspose->getOperator()->setDatatype(DataType::Int32);
+ myTranspose->getOperator()->setBackend("cpu");
+ myTranspose->getOperator()->associateInput(0,input);
+ myTranspose->getOperator()->computeOutputDims();
+ myTranspose->forward();
-}
-
-// TEST_CASE("[cpu/operator] Transpose(forward)") {
-// std::shared_ptr<Tensor> input = std::make_shared<Tensor>(Array3D<float,2,3,4> {
-// {
-// {{0.0, 0.1, 0.2, 0.3},
-// {0.4, 0.5, 0.6, 0.7},
-// {0.8, 0.9, 1.0, 1.1}},
-
-// {{1.2, 1.3, 1.4, 1.5},
-// {1.6, 1.7, 1.8, 1.9},
-// {2.0, 2.1, 2.2, 2.3}}
-// }
-// });
-// std::shared_ptr<Tensor> output = std::make_shared<Tensor>(Array3D<float,2,4,3> {
-// {
-// {{0.0, 0.4, 0.8},
-// {0.1, 0.5, 0.9},
-// {0.2, 0.6, 1.0},
-// {0.3, 0.7, 1.1}},
-
-// {{1.2, 1.6, 2.0},
-// {1.3, 1.7, 2.1},
-// {1.4, 1.8, 2.2},
-// {1.5, 1.9, 2.3}}
-// }
-// });
-// std::shared_ptr<Node> myTranspose = Transpose<3>(std::array<DimSize_t,3>{{0,2,1}});
-// myTranspose->getOperator()->setDatatype(DataType::Float32);
-// myTranspose->getOperator()->setBackend("cpu");
-// myTranspose->getOperator()->associateInput(0,input);
-// myTranspose->getOperator()->computeOutputDims();
-// myTranspose->forward();
-
-// float* resPtr = static_cast<float*>(myTranspose->getOperator()->getOutput(0)->getImpl()->rawPtr());
-// float* expectedPtr = static_cast<float*>(output->getImpl()->rawPtr());
-// for (std::size_t i = 0; i< 24; ++i) {
-// std::cout << "Res " << resPtr[i] << " , expected : " << expectedPtr[i] << std::endl;
-// REQUIRE(std::abs(resPtr[i]-expectedPtr[i]) < 0.00001);
-// }
-// // REQUIRE(*(myTranspose->getOperator()->getOutput(0)) == *output);
-
-// }
\ No newline at end of file
+ REQUIRE(*(myTranspose->getOperator()->getOutput(0)) == *output);
+ }
+}
\ No newline at end of file