GPU_1/run_cuda.cuh at main · Dudgit/GPU_1 · GitHub

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
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
// Include Cuda libraries, 'cause I use Visual Studio
#include "cuda_runtime.h"
#include "device_launch_parameters.h"
//Refference functions
#include "vector_mx_mul.h"
#include "cpp_functions.h"

__global__ void mx_vec_gpu(float* result_mx, float* input_vector, float* input_mx, int N)
{

    unsigned int y = blockIdx.y * blockDim.y + threadIdx.y;

    __shared__ float Atmp[MBS * MBS];
    __shared__ float Btmp[MBS * MBS];


    float sum = 0.f;

    for (int K = 0; K < N / MBS; ++K)
    {
        Atmp[threadIdx.y * MBS + threadIdx.x] = input_mx[y * N + (K * MBS + threadIdx.x)];
        Btmp[threadIdx.y * MBS + threadIdx.x] = input_vector[(K * MBS + threadIdx.y)];

        __syncthreads();
        for (int k = 0; k < MBS; ++k)
        {
            sum += Atmp[threadIdx.y * MBS + k] * Btmp[k * MBS + threadIdx.x];
        }
        __syncthreads();
    }
    result_mx[y] = sum;
}

float do_Cuda(std::vector<float>& A, std::vector<float>& B, std::vector<float>& C0, std::vector<float>& C1, std::vector<float>& C2)
{

    float* pA = nullptr;
    float* pB = nullptr;
    float* pC2 = nullptr;

    cudaEvent_t evt[2];
    for (auto& e : evt) { cudaEventCreate(&e); }

    cudaError_t err = cudaSuccess;

    err = cudaMalloc((void**)&pA, N * sizeof(float));
    if (err != cudaSuccess) { std::cout << "Error allocating CUDA memory: " << cudaGetErrorString(err) << "\n"; return -1; }

    err = cudaMalloc((void**)&pB, N * N * sizeof(float));
    if (err != cudaSuccess) { std::cout << "Error allocating CUDA memory: " << cudaGetErrorString(err) << "\n"; return -1; }

    err = cudaMalloc((void**)&pC2, N * sizeof(float));
    if (err != cudaSuccess) { std::cout << "Error allocating CUDA memory: " << cudaGetErrorString(err) << "\n"; return -1; }

    err = cudaMemcpy(pA, A.data(), N * sizeof(float), cudaMemcpyHostToDevice);
    if (err != cudaSuccess) { std::cout << "Error copying memory to device: " << cudaGetErrorString(err) << "\n"; return -1; }

    err = cudaMemcpy(pB, B.data(), N * N * sizeof(float), cudaMemcpyHostToDevice);
    if (err != cudaSuccess) { std::cout << "Error copying memory to device: " << cudaGetErrorString(err) << "\n"; return -1; }

    {
        dim3 dimGrid(n_blocks, n_blocks);
        dim3 dimBlock(block_sz, block_sz);
        cudaEventRecord(evt[0]);
        mx_vec_gpu << <dimGrid, dimBlock >> > (pC2, pA, pB, N);
        err = cudaGetLastError();
        if (err != cudaSuccess) { std::cout << "CUDA error in kernel call: " << cudaGetErrorString(err) << "\n"; return -1; }
        cudaEventRecord(evt[1]);
    }
    err = cudaMemcpy(C2.data(), pC2, N * sizeof(float), cudaMemcpyDeviceToHost);
    if (err != cudaSuccess) { std::cout << "Error copying memory to host: " << cudaGetErrorString(err) << "\n"; return -1; }

    err = cudaFree(pA);
    if (err != cudaSuccess) { std::cout << "Error freeing allocation: " << cudaGetErrorString(err) << "\n"; return -1; }

    err = cudaFree(pB);
    if (err != cudaSuccess) { std::cout << "Error freeing allocation: " << cudaGetErrorString(err) << "\n"; return -1; }

    err = cudaFree(pC2);
    if (err != cudaSuccess) { std::cout << "Error freeing allocation: " << cudaGetErrorString(err) << "\n"; return -1; }

    cudaEventSynchronize(evt[1]);
    float dt = 0.0f;//milliseconds
    cudaEventElapsedTime(&dt, evt[0], evt[1]);
    for (auto& e : evt) { cudaEventDestroy(e); }
    return dt;
}