-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathforward_gpu.cu~
183 lines (140 loc) · 4.81 KB
/
forward_gpu.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
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
179
180
181
182
183
/*
** Author: Tapas Kanungo, kanungo@cfar.umd.edu
** Date: 15 December 1997
** File: forward.c
** Purpose: Foward algorithm for computing the probabilty
** of observing a sequence given a HMM model parameter.
** Organization: University of Maryland
**
** $Id: forward.c,v 1.2 1998/02/19 12:42:31 kanungo Exp kanungo $
*/
#include <stdio.h>
#include "hmm.h"
#include "forward_kernel.cu"
static char rcsid[] = "$Id: forward.c,v 1.2 1998/02/19 12:42:31 kanungo Exp kanungo $";
//#ifdef __GPU
extern "C"
void ForwardGPU(HMM *phmm, int T, int *O, double **alpha, double *pprob)
{
int i, j; /* state indices */
int t; /* time index */
double sum; /* partial sum */
/* 1. Initialization */
for (i = 1; i <= phmm->N; i++)
alpha[1][i] = phmm->pi[i]* phmm->B[i][O[1]];
/// rearrange CPU memory
/// note that the time spent here won't be counted
int N = phmm->N;
int M = phmm->M;
real *h_A, *h_B, *h_alpha;
h_A = (real*)malloc(sizeof(real)*N*N);
h_B = (real*)malloc(sizeof(real)*N*M);
h_alpha = (real*)malloc(sizeof(real)*T*N);
for ( i=0; i<N; ++i )
{
for ( j=0; j<N; ++j)
h_A[j*N + i] = phmm->A[i+1][j+1];
for ( j=0; j<M; ++j )
h_B[j*N + i] = phmm->B[i+1][j+1];
}
for ( i=0; i<T; ++i )
for ( j=0; j<N; ++j )
{
h_alpha[i*N + j] = alpha[i+1][j+1];
}
/// timing starts from here, or later if you wish...
/// initialize the data on GPU
printf("\tRunning GPU accelerated version\n");
real *g_A, *g_B, *g_alpha;
/// it turned out that 2D memory allocation is problematic
cudaMalloc((void**)&g_A, sizeof(real)*N*N);
cudaMalloc((void**)&g_B, sizeof(real)*N*M);
cudaMalloc((void**)&g_alpha, sizeof(real)*N*T);
cudaMemcpy(g_A, h_A, sizeof(real)*N*N, cudaMemcpyHostToDevice);
cudaMemcpy(g_B, h_B, sizeof(real)*N*M, cudaMemcpyHostToDevice);
cudaMemcpy(g_alpha, h_alpha, sizeof(real)*N*T, cudaMemcpyHostToDevice);
/// function signature
for (t=1; t<T; ++t)
{
//ForwardKernelv1<<<N/32, 32>>>(O[t+1], (g_delta + (t-1)*N), (g_delta + t*N), (g_psi + t*N), g_A, g_B, N);
ForwardKernel<<<N/32, 32>>>(O[t+1], (g_alpha + (t-1)*N), (g_alpha + t*N), g_A, g_B, N);
}
cudaMemcpy(h_alpha, g_alpha, sizeof(real)*N*T, cudaMemcpyDeviceToHost);
/* /\* 2. Induction *\/ */
/* for (t = 1; t < T; t++) */
/* { */
/* for (j = 1; j <= phmm->N; j++) */
/* { */
/* sum = 0.0; */
/* /// TODO: transpose A */
/* /// this is a dot product, consider MKL */
/* for (i = 1; i <= phmm->N; i++) */
/* sum += alpha[t][i]* (phmm->A[i][j]); */
/* alpha[t+1][j] = sum*(phmm->B[j][O[t+1]]); */
/* } */
/* } */
/* 3. Termination */
*pprob = 0.0;
for (i = 1; i <= phmm->N; i++)
//*pprob += alpha[T][i];
*pprob += h_alpha[(T-1)*N + i-1];
}
//#else
/* void Forward(HMM *phmm, int T, int *O, double **alpha, double *pprob) */
/* { */
/* int i, j; /\* state indices *\/ */
/* int t; /\* time index *\/ */
/* double sum; /\* partial sum *\/ */
/* /\* 1. Initialization *\/ */
/* for (i = 1; i <= phmm->N; i++) */
/* alpha[1][i] = phmm->pi[i]* phmm->B[i][O[1]]; */
/* /\* 2. Induction *\/ */
/* for (t = 1; t < T; t++) { */
/* for (j = 1; j <= phmm->N; j++) { */
/* sum = 0.0; */
/* /// TODO: transpose A */
/* /// this is a dot product, consider MKL */
/* for (i = 1; i <= phmm->N; i++) */
/* sum += alpha[t][i]* (phmm->A[i][j]); */
/* alpha[t+1][j] = sum*(phmm->B[j][O[t+1]]); */
/* } */
/* } */
/* /\* 3. Termination *\/ */
/* *pprob = 0.0; */
/* for (i = 1; i <= phmm->N; i++) */
/* *pprob += alpha[T][i]; */
/* } */
//#endif
/* void ForwardWithScale(HMM *phmm, int T, int *O, double **alpha, */
/* double *scale, double *pprob) */
/* /\* pprob is the LOG probability *\/ */
/* { */
/* int i, j; /\* state indices *\/ */
/* int t; /\* time index *\/ */
/* double sum; /\* partial sum *\/ */
/* /\* 1. Initialization *\/ */
/* scale[1] = 0.0; */
/* for (i = 1; i <= phmm->N; i++) { */
/* alpha[1][i] = phmm->pi[i]* (phmm->B[i][O[1]]); */
/* scale[1] += alpha[1][i]; */
/* } */
/* for (i = 1; i <= phmm->N; i++) */
/* alpha[1][i] /= scale[1]; */
/* /\* 2. Induction *\/ */
/* for (t = 1; t <= T - 1; t++) { */
/* scale[t+1] = 0.0; */
/* for (j = 1; j <= phmm->N; j++) { */
/* sum = 0.0; */
/* for (i = 1; i <= phmm->N; i++) */
/* sum += alpha[t][i]* (phmm->A[i][j]); */
/* alpha[t+1][j] = sum*(phmm->B[j][O[t+1]]); */
/* scale[t+1] += alpha[t+1][j]; */
/* } */
/* for (j = 1; j <= phmm->N; j++) */
/* alpha[t+1][j] /= scale[t+1]; */
/* } */
/* /\* 3. Termination *\/ */
/* *pprob = 0.0; */
/* for (t = 1; t <= T; t++) */
/* *pprob += log(scale[t]); */
/* } */