Skip to content

GitLab

"...amoeba/platforms/cuda2/include/AmoebaCudaKernelFactory.h" did not exist on "0e879806cdd38e58b04481ecf7fcd93c44c7dc27"
Switch branch/tag
Find file Normal viewHistoryPermalink
removeCM.cl 3.02 KB
Newer Older
Peter Eastman's avatar
Created OpenCL implementation of CMMotionRemover. Also made a few optimizations.  
Peter Eastman committed
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 
/**
 * Calculate the center of mass momentum.
 */

__kernel void calcCenterOfMassMomentum(int numAtoms, __global float4* velm, __global float4* cmMomentum, __local float4* temp) {
    int index = get_global_id(0);
    float4 cm = 0.0f;
    while (index < numAtoms) {
        float4 velocity = velm[index];
        cm.xyz += velocity.xyz/velocity.w;
        index += get_global_size(0);
    }

    // Sum the threads in this group.

    int thread = get_local_id(0);
    temp[thread] = cm;
    barrier(CLK_LOCAL_MEM_FENCE);
#ifdef WARPS_ARE_ATOMIC
    if (thread < 32) {
        temp[thread] += temp[thread+32];
        if (thread < 16)
            temp[thread] += temp[thread+16];
        if (thread < 8)
            temp[thread] += temp[thread+8];
        if (thread < 4)
            temp[thread] += temp[thread+4];
        if (thread < 2)
            temp[thread] += temp[thread+2];
    }
#else
    if (thread < 32)
        temp[thread] += temp[thread+32];
    barrier(CLK_LOCAL_MEM_FENCE);
    if (thread < 16)
        temp[thread] += temp[thread+16];
    barrier(CLK_LOCAL_MEM_FENCE);
    if (thread < 8)
        temp[thread] += temp[thread+8];
    barrier(CLK_LOCAL_MEM_FENCE);
    if (thread < 4)
        temp[thread] += temp[thread+4];
    barrier(CLK_LOCAL_MEM_FENCE);
    if (thread < 2)
        temp[thread] += temp[thread+2];
    barrier(CLK_LOCAL_MEM_FENCE);
#endif
    if (thread == 0)
        cmMomentum[get_group_id(0)] = temp[thread]+temp[thread+1];
}

/**
 * Remove center of mass motion.
 */

__kernel void removeCenterOfMassMomentum(int numAtoms, __global float4* velm, __global float4* cmMomentum, __local float4* temp) {
    // First sum all of the momenta that were calculated by individual groups.

    int index = get_local_id(0);
    float4 cm = 0.0f;
    while (index < get_num_groups(0)) {
        cm += cmMomentum[index];
        index += get_local_size(0);
    }
    int thread = get_local_id(0);
    temp[thread] = cm;
    barrier(CLK_LOCAL_MEM_FENCE);
#ifdef WARPS_ARE_ATOMIC
    if (thread < 32) {
        temp[thread] += temp[thread+32];
        if (thread < 16)
            temp[thread] += temp[thread+16];
        if (thread < 8)
            temp[thread] += temp[thread+8];
        if (thread < 4)
            temp[thread] += temp[thread+4];
        if (thread < 2)
            temp[thread] += temp[thread+2];
    }
#else
    if (thread < 32)
        temp[thread] += temp[thread+32];
    barrier(CLK_LOCAL_MEM_FENCE);
    if (thread < 16)
        temp[thread] += temp[thread+16];
    barrier(CLK_LOCAL_MEM_FENCE);
    if (thread < 8)
        temp[thread] += temp[thread+8];
    barrier(CLK_LOCAL_MEM_FENCE);
    if (thread < 4)
        temp[thread] += temp[thread+4];
    barrier(CLK_LOCAL_MEM_FENCE);
    if (thread < 2)
        temp[thread] += temp[thread+2];
#endif
Peter Eastman's avatar
Fixed threading bug in CM motion removal  
Peter Eastman committed
96 
    barrier(CLK_LOCAL_MEM_FENCE);
Peter Eastman's avatar
Created OpenCL implementation of CMMotionRemover. Also made a few optimizations.  
Peter Eastman committed
97
98
99
100
101
102
103
104
105
106 
    cm = (temp[0]+temp[1])*INVERSE_TOTAL_MASS;

    // Now remove the center of mass velocity from each atom.

    index = get_global_id(0);
    while (index < numAtoms) {
        velm[index].xyz -= cm.xyz;
        index += get_global_size(0);
    }
}