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
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
#include<stdio.h>
#include<stdlib.h>

#define TILE_W 32
#define MAX_BLOCK 32

int getSize(char* fileName);
void writeMatrix(char* fileName, int SIZE, float* matrixToSave);
float* readMatrix(char* fileName);

__global__ void matMultiply(float *Ad, float *Bd, float *AxBd, int SIZE, int iter){
        int i = threadIdx.y + blockIdx.y*TILE_W;
        int j = threadIdx.x + blockIdx.x*TILE_W;
        int k;
        float AxB_ij;
        if(iter == 0){
                for(k=0; k<SIZE; k++){
                        AxB_ij += Ad[i*SIZE + k] * Bd[k*SIZE +j];
                }
                AxBd[i*SIZE +j] = AxB_ij;
        }
        else {
                while (i<SIZE){
                        while (j<SIZE){
                                AxB_ij = 0;
                                for(k=0; k<SIZE; k++){
                                        AxB_ij += Ad[i*SIZE+k] * Bd[k*SIZE+j];
                                }
                                AxBd[i*SIZE +j] = AxB_ij;
                                j += blockDim.x * gridDim.x;
                        }
                        i += blockDim.y * gridDim.y;
                }
        }
}


int main(){
        float *A_dev, *B_dev, *AxB_dev;
        int memSize;
        int blockSize;
        float ratio;
        int residue;
        int kerIter;
        int thSize;

        float *A, *B, *AxB;
        int SIZE_A, SIZE_B;

        A = (float*) readMatrix("A.mat");
        SIZE_A = (int) getSize("A.mat");

        B = (float*) readMatrix("B.mat");
        SIZE_B = (int) getSize("B.mat");

        if(SIZE_A != SIZE_B){
                printf("size not match\n");
                exit(-1);
        }

        AxB = (float*) malloc(SIZE_A*SIZE_A * sizeof(float));
        if(AxB ==NULL){
                free(AxB);
                printf("malloc failed\n");
                exit(-1);
        }

        memSize = SIZE_A*SIZE_A * sizeof(float);

        //prepare memory and values on GPU
        cudaMalloc((void**) &A_dev, memSize);
        cudaMemcpy(A_dev, A, memSize, cudaMemcpyHostToDevice);

        cudaMalloc((void**) &B_dev, memSize);
        cudaMemcpy(B_dev, B, memSize, cudaMemcpyHostToDevice);

        cudaMalloc((void**) &AxB_dev, memSize);

        //ditermin the grid and block dimensions
        ratio = SIZE_A / TILE_W;
        residue = SIZE_A % TILE_W;
        if (ratio<MAX_BLOCK){
                if(residue==0) blockSize = (int) ratio;
                else blockSize = ((int) (SIZE_A / TILE_W)) + 1;
                kerIter = 0;
        }else{
                blockSize = MAX_BLOCK;
                kerIter = 1;
        }
        dim3 block(blockSize, blockSize);

        if(TILE_W > SIZE_A) thSize = SIZE_A;
        else thSize = TILE_W;
        dim3 threads(thSize, thSize);


//***   kernel function call
        matMultiply<<<block, threads>>>(A_dev, B_dev, AxB_dev, SIZE_A, kerIter);

//***


        //transfer the GPU output to Host
        cudaMemcpy(AxB, AxB_dev, memSize, cudaMemcpyDeviceToHost);

        writeMatrix("multTest_CU.mat", SIZE_A, AxB);;
        free(A);
        free(B);
        free(AxB);

        //deallocat the cuda memory
        cudaFree(A_dev);
        cudaFree(B_dev);
        cudaFree(AxB_dev);
}

float* readMatrix(char* fileName);
int getSize(char* fileName);
void writeMatrix(char* fileName, int SIZE, float* matrixToSave);

float* readMatrix(char* fileName){
        int i, j, ni;
        int SIZE;
        float* matrix;
        FILE *mfPtr;

        if((mfPtr = fopen(fileName, "r")) == NULL)
                printf("File could not be opened\n");
        else{
                fscanf(mfPtr, "%d", &SIZE);
                matrix = (float*) malloc(SIZE*SIZE * sizeof(float));
                if(NULL == matrix){
                        free(matrix);
                        printf("malloc failed\n");
                        exit(-1);
                }
                for(i=0; i<SIZE; i++){
                        ni = i*SIZE;
                        for(j=0; j<SIZE; j++){
                                fscanf(mfPtr, "%f", &matrix[ni+j]);
                        }
                }
                fclose(mfPtr);
        }
        return matrix;
}

void writeMatrix(char* fileName, int SIZE, float* matrixToSave){
        int i, j, ni;
        FILE *mfPtr;
        if((mfPtr = fopen(fileName, "w")) == NULL){
                printf("File could not be opened\n");
        }
        else{
                fprintf(mfPtr, "%d\n", SIZE);
                printf("%s is saved\n", fileName);
                printf("(size of %d by %d)\n", SIZE, SIZE);
                for (i=0; i<SIZE; i++){
                        ni = i*SIZE;
                        for (j=0; j<SIZE; j++){
                                fprintf(mfPtr, "%4.4f\n", matrixToSave[ni+j]);
                        }
                }
                fclose(mfPtr);
        }
}

int getSize(char* fileName){
        int SIZE;
        FILE *mfPtr;

        if((mfPtr = fopen(fileName, "r")) == NULL)
                printf("File could not be opened\n");
        else{
                fscanf(mfPtr, "%d", &SIZE);
        }
        return SIZE;
}