Mathis/CUDANet
1
0
Fork 0
mirror of https://github.com/lordmathis/CUDANet.git synced 2025-12-22 22:34:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

WIP Migrate Dense layer

Browse Source
This commit is contained in:
Matúš Námešný
2025-11-18 21:12:47 +01:00
parent 64eac7050b
commit 7f203b8947
14 changed files with 116 additions and 221 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
include/backend.hpp
View File

@@ -4,27 +4,41 @@
#include "tensor.hpp"
namespace CUDANet
{
class Backend
{
public:
namespace CUDANet {
class Backend {
public:
// Memory management
virtual void* allocate(size_t bytes) = 0;
virtual void deallocate(void* ptr) = 0;
virtual void deallocate(void* ptr) = 0;
// Tensor ops
virtual void print(const CUDANet::Tensor &input) = 0;
virtual void zero(CUDANet::Tensor &input) = 0;
virtual void sum(const CUDANet::Tensor &input, CUDANet::Tensor &sum) = 0;
virtual void max(const CUDANet::Tensor &input, CUDANet::Tensor &max) = 0;
virtual void print(const CUDANet::Tensor& input) = 0;
virtual void zero(CUDANet::Tensor& input) = 0;
virtual void
copy_to_device(CUDANet::Tensor& tensor, void* data, size_t size) = 0;
virtual void sum(const CUDANet::Tensor& input, CUDANet::Tensor& sum) = 0;
virtual void max(const CUDANet::Tensor& input, CUDANet::Tensor& max) = 0;
// Layer ops
virtual void relu(CUDANet::Tensor &tensor) = 0;
virtual void sigmoid(CUDANet::Tensor &tensor) = 0;
virtual void softmax(CUDANet::Tensor &tensor, CUDANet::Tensor &temp_max, CUDANet::Tensor &temp_sum) = 0;
virtual void relu(CUDANet::Tensor& tensor) = 0;
virtual void sigmoid(CUDANet::Tensor& tensor) = 0;
virtual void softmax(
CUDANet::Tensor& tensor,
CUDANet::Tensor& temp_max,
CUDANet::Tensor& temp_sum
) = 0;
virtual CUDANet::Tensor& dense(
CUDANet::Tensor& weights,
CUDANet::Tensor& biases,
CUDANet::Tensor& input,
CUDANet::Tensor& output,
size_t input_size,
size_t output_size
) = 0;
};
} // namespace CUDANet::Backend
} // namespace CUDANet

33
include/backend/cuda.cuh
View File

@@ -6,21 +6,36 @@
namespace CUDANet::Backend {
class CUDA : public Backend {
public:
public:
// Memory management
void* allocate(size_t bytes) override;
void deallocate(void* ptr) override;
void deallocate(void* ptr) override;
// Tensor ops
void print(const CUDANet::Tensor &input) override;
void zero(CUDANet::Tensor &input) override;
void sum(const CUDANet::Tensor &input, CUDANet::Tensor &sum) override;
void max(const CUDANet::Tensor &input, CUDANet::Tensor &max) override;
void print(const CUDANet::Tensor& input) override;
void zero(CUDANet::Tensor& input) override;
void
copy_to_device(CUDANet::Tensor& tensor, void* data, size_t size) override;
void sum(const CUDANet::Tensor& input, CUDANet::Tensor& sum) override;
void max(const CUDANet::Tensor& input, CUDANet::Tensor& max) override;
// Layer ops
void relu(CUDANet::Tensor &tensor) override;
void sigmoid(CUDANet::Tensor &tensor) override;
void softmax(CUDANet::Tensor &tensor, CUDANet::Tensor &temp_max, CUDANet::Tensor &temp_sum) override;
void relu(CUDANet::Tensor& tensor) override;
void sigmoid(CUDANet::Tensor& tensor) override;
void softmax(
CUDANet::Tensor& tensor,
CUDANet::Tensor& temp_max,
CUDANet::Tensor& temp_sum
) override;
CUDANet::Tensor& dense(
CUDANet::Tensor& weights,
CUDANet::Tensor& biases,
CUDANet::Tensor& input,
CUDANet::Tensor& output,
size_t input_size,
size_t output_size
) override;
};
} // namespace CUDANet::Backend

4
include/layer.hpp
View File

@@ -30,11 +30,11 @@ class Layer {
virtual size_t output_size() = 0;
virtual void set_weights(CUDANet::Tensor &input) = 0;
virtual void set_weights(void *input) = 0;
virtual CUDANet::Tensor& get_weights() = 0;
virtual void set_biases(CUDANet::Tensor &input) = 0;
virtual void set_biases(void *input) = 0;
virtual CUDANet::Tensor& get_biases() = 0;
};

18
include/layers/activation.hpp
View File

@@ -29,23 +29,23 @@ class Activation : public Layer {
~Activation() = default;
CUDANet::Tensor& forward(CUDANet::Tensor &input);
CUDANet::Tensor& forward(CUDANet::Tensor &input) override;
CUDANet::Shape input_shape();
CUDANet::Shape input_shape() override;
CUDANet::Shape output_shape();
CUDANet::Shape output_shape() override;
size_t input_size();
size_t input_size() override;
size_t output_size();
size_t output_size() override;
void set_weights(CUDANet::Tensor &input);
void set_weights(void *input) override;
CUDANet::Tensor& get_weights();
CUDANet::Tensor& get_weights() override;
void set_biases(CUDANet::Tensor &input);
void set_biases(void *input) override;
CUDANet::Tensor& get_biases();
CUDANet::Tensor& get_biases() override;
private:

45
include/layers/dense.hpp
View File

@@ -18,23 +18,23 @@ class Dense : public Layer {
~Dense();
CUDANet::Tensor& forward(CUDANet::Tensor &input);
CUDANet::Tensor& forward(CUDANet::Tensor &input) override;
CUDANet::Shape input_shape();
CUDANet::Shape input_shape() override;
CUDANet::Shape output_shape();
CUDANet::Shape output_shape() override;
size_t input_size();
size_t input_size() override;
size_t output_size();
size_t output_size() override;
void set_weights(CUDANet::Tensor &input);
void set_weights(void *input) override;
CUDANet::Tensor& get_weights();
CUDANet::Tensor& get_weights() override;
void set_biases(CUDANet::Tensor &input);
void set_biases(void *input) override;
CUDANet::Tensor& get_biases();
CUDANet::Tensor& get_biases() override;
private:
CUDANet::Backend *backend;
@@ -45,32 +45,7 @@ class Dense : public Layer {
CUDANet::Tensor weights;
CUDANet::Tensor biases;
void init_weights();
void init_biases();
// #ifdef USE_CUDA
// float* d_output;
// float* d_weights;
// float* d_biases;
// // Precompute kernel launch parameters
// int forwardGridSize;
// int biasGridSize;
// /**
// * @brief Copy the weights and biases to the device
// *
// */
// void toCuda();
// void initCUDA();
// void delCUDA();
// float* forwardCUDA(const float* d_input);
// #endif
CUDANet::Tensor output;
};
} // namespace CUDANet::Layers