matrix_divide_scalar.cu File Reference

Implementation of the Matrix::divide_scalar method for GPU-accelerated division of a matrix by a scalar. More...

#include "matrix.h"
#include <cuda_runtime.h>
#include <stdexcept>
#include <cfloat>
#include <cmath>

Include dependency graph for matrix_divide_scalar.cu:

Go to the source code of this file.

Functions
__global__ void	divideScalarKernel (double *data, double scalar, int size)
	CUDA kernel for dividing matrix elements by a scalar. More...

Detailed Description

Implementation of the Matrix::divide_scalar method for GPU-accelerated division of a matrix by a scalar.

Definition in file matrix_divide_scalar.cu.

Function Documentation

divideScalarKernel()

__global__ void divideScalarKernel	(	double *	data,
		double	scalar,
		int	size
	)

CUDA kernel for dividing matrix elements by a scalar.

Parameters

data	Pointer to the matrix data.
scalar	The scalar to divide by.
size	Total number of elements in the matrix.

Definition at line 18 of file matrix_divide_scalar.cu.

                                                                          {
    // Calculate global thread index
    int idx = blockIdx.x * blockDim.x + threadIdx.x;
 
    // Check if thread is within matrix bounds
    if (idx < size) {
        // Handle division by very small numbers
        if (fabs(scalar) < DBL_EPSILON) {
            // If data is zero, keep it zero
            // Otherwise, set to max or min based on sign
            data[idx] = (data[idx] == 0.0) ? 0.0 : ((data[idx] > 0.0) ? DBL_MAX : -DBL_MAX);
        } 
        // Handle very large numbers
        else if (fabs(data[idx]) > DBL_MAX / 2) {
            // Preserve sign and set to max value
            data[idx] = (data[idx] > 0.0) ? DBL_MAX : -DBL_MAX;
        } 
        // Regular division for normal cases
        else {
            data[idx] /= scalar;
        }
    }
}