kernels.cuh File Reference

#include "common.cuh"
#include "cublas_handle.cuh"
#include <cmath>

Include dependency graph for kernels.cuh:

This graph shows which files directly or indirectly include this file:

Namespaces
	cuda_mlp

Enumerations
enum class	cuda_mlp::ActivationType : int { cuda_mlp::Linear = 0 , cuda_mlp::Tanh = 1 , cuda_mlp::ReLU = 2 , cuda_mlp::Sigmoid = 3 }
	Supported activation functions. More...

Functions
void	cuda_mlp::device_set_zero (CudaScalar *ptr, size_t n)
	Set device memory to zero. More...
void	cuda_mlp::device_copy (CudaScalar *dst, const CudaScalar *src, size_t n)
	Copy device-to-device. More...
CudaScalar	cuda_mlp::device_dot (CublasHandle &handle, const CudaScalar *x, const CudaScalar *y, int n)
	Compute dot product on device using cuBLAS. More...
CudaScalar	cuda_mlp::device_nrm2 (CublasHandle &handle, const CudaScalar *x, int n)
	Compute Euclidean norm on device using cuBLAS. More...
void	cuda_mlp::device_axpy (CublasHandle &handle, int n, CudaScalar alpha, const CudaScalar *x, CudaScalar *y)
	y <- alpha * x + y (AXPY) on device using cuBLAS. More...
void	cuda_mlp::device_scal (CublasHandle &handle, int n, CudaScalar alpha, CudaScalar *x)
	Scale vector x <- alpha * x on device using cuBLAS. More...
CudaScalar	cuda_mlp::activation_scale (ActivationType act)
	scaling factor for initialization. More...
__global__ void	cuda_mlp::add_bias_kernel (CudaScalar *z, const CudaScalar *b, int rows, int cols)
	Kernel: add bias vector to column-major matrix. More...
__global__ void	cuda_mlp::activation_kernel (CudaScalar *a, int n, int act)
	Kernel: apply activation in-place. More...
__global__ void	cuda_mlp::activation_deriv_kernel (CudaScalar *grad, const CudaScalar *a, int n, int act)
	Kernel: multiply gradient by activation derivative. More...
__global__ void	cuda_mlp::diff_kernel (const CudaScalar *output, const CudaScalar *target, CudaScalar *diff, int n)
	Kernel: diff = output - target. More...
__global__ void	cuda_mlp::sum_rows_kernel (const CudaScalar *mat, CudaScalar *out, int rows, int cols)
	Kernel: sum columns (rows x cols) into a row vector. More...
void	cuda_mlp::launch_add_bias (CudaScalar *z, const CudaScalar *b, int rows, int cols)
	Launch add-bias kernel. More...
void	cuda_mlp::launch_activation (CudaScalar *a, int n, ActivationType act)
	Launch activation kernel. More...
void	cuda_mlp::launch_activation_deriv (CudaScalar *grad, const CudaScalar *a, int n, ActivationType act)
	Launch activation-derivative kernel. More...
void	cuda_mlp::launch_diff (const CudaScalar *output, const CudaScalar *target, CudaScalar *diff, int n)
	Launch diff kernel. More...
void	cuda_mlp::launch_sum_rows (const CudaScalar *mat, CudaScalar *out, int rows, int cols)
	Launch sum-rows kernel. More...