-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathlbm_chapter5_7_D2Q9_cuda.cu
145 lines (132 loc) · 4.66 KB
/
lbm_chapter5_7_D2Q9_cuda.cu
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
#include <cuda_runtime.h>
#include <device_launch_parameters.h>
#include <iostream>
#define m 256
#define n 256
#define alpha 0.25f
#define omega (1.0 / (3.0 * alpha + 0.5))
#define twall 1.0f
#define nstep 2000
__global__ void collision(float* f_old, float* f, float* w, float* rho) {
int x = threadIdx.x + blockIdx.x * blockDim.x;
int y = threadIdx.y + blockIdx.y * blockDim.y;
int k = blockIdx.z;
f_old[y * m * 9 + x * 9 + k] = (1 - omega) * f[y * m * 9 + x * 9 + k] + omega * w[k] * rho[y * m + x];
}
__global__ void stream(float* f_old, float* f, float* e) {
int x = threadIdx.x + blockIdx.x * blockDim.x;
int y = threadIdx.y + blockIdx.y * blockDim.y;
int k = blockIdx.z;
int xp = x - e[k * 2 + 0];
int yp = y - e[k * 2 + 1];
if (xp >= 0 && xp < m && yp >= 0 && yp < n) {
f[y * m * 9 + x * 9 + k] = f_old[yp * m * 9 + xp * 9 + k];
}
}
__global__ void boundary_tb(float* f) {
int x = threadIdx.x + blockIdx.x * blockDim.x;
f[0 * m * 9 + x * 9 + 2] = f[1 * m * 9 + x * 9 + 2];
f[0 * m * 9 + x * 9 + 5] = f[1 * m * 9 + x * 9 + 5];
f[0 * m * 9 + x * 9 + 6] = f[1 * m * 9 + x * 9 + 6];
f[(n - 1) * m * 9 + x * 9 + 7] = -f[(n - 1) * m * 9 + x * 9 + 5];
f[(n - 1) * m * 9 + x * 9 + 4] = -f[(n - 1) * m * 9 + x * 9 + 2];
f[(n - 1) * m * 9 + x * 9 + 8] = -f[(n - 1) * m * 9 + x * 9 + 6];
}
__global__ void boundary_lr(float* f, float* w) {
int y = threadIdx.x + blockIdx.x * blockDim.x;
f[y * m * 9 + 0 * 9 + 1] = w[1] * twall + w[3] * twall - f[y * m * 9 + 0 * 9 + 3];
f[y * m * 9 + 0 * 9 + 5] = w[5] * twall + w[7] * twall - f[y * m * 9 + 0 * 9 + 7];
f[y * m * 9 + 0 * 9 + 8] = w[8] * twall + w[6] * twall - f[y * m * 9 + 0 * 9 + 6];
f[y * m * 9 + (m - 1) * 9 + 3] = -f[y * m * 9 + (m - 1) * 9 + 1];
f[y * m * 9 + (m - 1) * 9 + 7] = -f[y * m * 9 + (m - 1) * 9 + 5];
f[y * m * 9 + (m - 1) * 9 + 6] = -f[y * m * 9 + (m - 1) * 9 + 8];
}
__global__ void update(float* rho, float* f) {
int x = threadIdx.x + blockIdx.x * blockDim.x;
int y = threadIdx.y + blockIdx.y * blockDim.y;
rho[y * m + x] = f[y * m * 9 + x * 9 + 0] + f[y * m * 9 + x * 9 + 1] + f[y * m * 9 + x * 9 + 2] + f[y * m * 9 + x * 9 + 3] + f[y * m * 9 + x * 9 + 4] + f[y * m * 9 + x * 9 + 5] + f[y * m * 9 + x * 9 + 6] + f[y * m * 9 + x * 9 + 7] + f[y * m * 9 + x * 9 + 8];
}
class lbm_solver {
float w[9] = { 4. / 9., 1. / 9., 1. / 9., 1. / 9., 1. / 9.
, 1. / 36., 1. / 36., 1. / 36., 1. / 36. };
float e[9 * 2] = { 0, 0, 1, 0, 0, 1, -1, 0, 0, -1, 1, 1,
-1, 1, -1, -1, 1, -1 };
float* rho, * f, * f_old;
float* dev_rho, * dev_f, * dev_f_old, * dev_w, * dev_e;
float Tm[m];
public:
lbm_solver() {
rho = new float[m * n];
f = new float[m * n * 9];
f_old = new float[m * n * 9];
for (int x = 0; x < m; x++) {
for (int y = 0; y < n; y++) {
rho[x * n + y] = 0;
for (int k = 0; k < 9; k++) {
f[x * n * 9 + y * 9 + k] = 0;
f_old[y * m * 9 + x * 9 + k] = 0;
}
}
}
cudaMalloc((void**)&dev_rho, sizeof(float) * m * n);
cudaMalloc((void**)&dev_f, sizeof(float) * m * n * 9);
cudaMalloc((void**)&dev_f_old, sizeof(float) * m * n * 9);
cudaMalloc((void**)&dev_w, sizeof(float) * 9);
cudaMalloc((void**)&dev_e, sizeof(float) * 9 * 2);
cudaMemcpy(dev_rho, rho, sizeof(float) * m * n, cudaMemcpyHostToDevice);
cudaMemcpy(dev_f, f, sizeof(float) * m * n * 9, cudaMemcpyHostToDevice);
cudaMemcpy(dev_f_old, f_old, sizeof(float) * m * n * 9, cudaMemcpyHostToDevice);
cudaMemcpy(dev_w, w, sizeof(float) * 9, cudaMemcpyHostToDevice);
cudaMemcpy(dev_e, e, sizeof(float) * 9 * 2, cudaMemcpyHostToDevice);
}
~lbm_solver() {
cudaFree(dev_rho);
cudaFree(dev_f);
cudaFree(dev_f_old);
cudaFree(dev_e);
cudaFree(dev_w);
delete[] f;
delete[] f_old;
delete[] rho;
}
void post() {
cudaMemcpy(rho, dev_rho, sizeof(float) * m * n, cudaMemcpyDeviceToHost);
for (int x = 0; x < m; x++) {
Tm[x] = rho[(n - 1) / 2 * m + x];
}
}
void solve() {
dim3 grid2d(m / 16, n / 16);
dim3 grid3d(m / 16, n / 16, 9);
dim3 threads(16, 16);
for (int k = 0; k < nstep; k++) {
collision << <grid3d, threads >> > (dev_f_old, dev_f, dev_w, dev_rho);
cudaDeviceSynchronize();
stream << <grid3d, threads >> > (dev_f_old, dev_f, dev_e);
cudaDeviceSynchronize();
boundary_tb << <n / 16, 16 >> > (dev_f);
cudaDeviceSynchronize();
boundary_lr << <m / 16, 16 >> > (dev_f, dev_w);
cudaDeviceSynchronize();
update << <grid2d, threads >> > (dev_rho, dev_f);
cudaDeviceSynchronize();
}
post();
}
void plot() {
/*for (int y = 0; y < m; y++) {
for (int x = 0; x < 2; x++) {
std::cout << rho[y * m + x] << ",";
}
std::cout << std::endl;
}*/
for (int x = 0; x < m; x++) {
std::cout << Tm[x] << std::endl;
}
}
};
int main() {
lbm_solver lbm;
lbm.solve();
lbm.plot();
}