matrix_argmax.cu File Reference

GPU implementation of the column-wise argmax function for matrices. More...

#include "matrix.h"
#include <cuda_runtime.h>

Include dependency graph for matrix_argmax.cu:

Go to the source code of this file.

Functions
__global__ void	argmax_GPU (const double *m, double *result, int rows, int cols)
	CUDA kernel for computing the argmax of each column in a matrix. More...

Detailed Description

GPU implementation of the column-wise argmax function for matrices.

Definition in file matrix_argmax.cu.

Function Documentation

argmax_GPU()

__global__ void argmax_GPU	(	const double *	m,
		double *	result,
		int	rows,
		int	cols
	)

CUDA kernel for computing the argmax of each column in a matrix.

Parameters

m	Pointer to the matrix data on the GPU.
result	Pointer to the result vector on the GPU.
rows	Number of rows in the matrix.
cols	Number of columns in the matrix.

Definition at line 16 of file matrix_argmax.cu.

                                                                                {
    // Determine the column this thread is responsible for
    int col = blockIdx.x * blockDim.x + threadIdx.x;
 
    // Proceed if the column index is within matrix bounds
    if (col < cols) {
        // Initialize max_val with the first element in the column and max_idx to the first row
        double max_val = m[col];
        int max_idx = 0;
 
        // Iterate through the rows to find the maximum value in the column
        for (int row = 1; row < rows; row++) {
            double val = m[row * cols + col]; // Access element (row, col)
            if (val > max_val) {
                max_val = val;
                max_idx = row;
            }
        }
        
        // Store the index of the maximum value in the result vector for this column
        result[col] = static_cast<double>(max_idx);
    }
}