array_cos.cuh
1.44 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
#ifndef STIM_CUDA_ARRAY_COS_H
#define STIM_CUDA_ARRAY_COS_H
#include <iostream>
#include <cuda.h>
#include <cmath>
#include <stim/cuda/cudatools.h>
namespace stim{
namespace cuda{
template<typename T>
__global__ void cuda_cos(T* ptr1, T* out, unsigned int N){
//calculate the 1D index for this thread
int idx = blockIdx.x * blockDim.x + threadIdx.x;
if(idx < N){
out[idx] = cos(ptr1[idx]);
}
}
template<typename T>
void gpu_cos(T* ptr1, T* out, unsigned int N){
//get the maximum number of threads per block for the CUDA device
int threads = stim::maxThreadsPerBlock();
//calculate the number of blocks
int blocks = N / threads + 1;
//call the kernel to do the multiplication
cuda_cos <<< blocks, threads >>>(ptr1, out, N);
}
template<typename T>
void cpu_cos(T* ptr1, T* cpu_out, unsigned int N){
//allocate memory on the GPU for the array
T* gpu_ptr1;
T* gpu_out;
HANDLE_ERROR( cudaMalloc( &gpu_ptr1, N * sizeof(T) ) );
HANDLE_ERROR( cudaMalloc( &gpu_out, N * sizeof(T) ) );
//copy the array to the GPU
HANDLE_ERROR( cudaMemcpy( gpu_ptr1, ptr1, N * sizeof(T), cudaMemcpyHostToDevice) );
//call the GPU version of this function
gpu_cos<T>(gpu_ptr1 ,gpu_out, N);
//copy the array back to the CPU
HANDLE_ERROR( cudaMemcpy( cpu_out, gpu_out, N * sizeof(T), cudaMemcpyDeviceToHost) );
//free allocated memory
cudaFree(gpu_ptr1);
cudaFree(gpu_out);
}
}
}
#endif