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
|
/*
* diffraction.cl
*
* GPU calculation kernel for truncated lattice diffraction
*
* (c) 2006-2010 Thomas White <taw@physics.org>
*
* Part of CrystFEL - crystallography with a FEL
*
*/
#define INDMAX 70
#define IDIM (INDMAX*2 +1)
float4 quat_rot(float4 q, float4 z)
{
float4 res;
float t01, t02, t03, t11, t12, t13, t22, t23, t33;
t01 = z.x*z.y;
t02 = z.x*z.z;
t03 = z.x*z.w;
t11 = z.y*z.y;
t12 = z.y*z.z;
t13 = z.y*z.w;
t22 = z.z*z.z;
t23 = z.z*z.w;
t33 = z.w*z.w;
res.x = (1.0 - 2.0 * (t22 + t33)) * q.x
+ (2.0 * (t12 + t03)) * q.y
+ (2.0 * (t13 - t02)) * q.z;
res.y = (2.0 * (t12 - t03)) * q.x
+ (1.0 - 2.0 * (t11 + t33)) * q.y
+ (2.0 * (t01 + t23)) * q.z;
res.z = (2.0 * (t02 + t13)) * q.x
+ (2.0 * (t23 - t01)) * q.y
+ (1.0 - 2.0 * (t11 + t22)) * q.z;
return res;
}
float4 get_q(int x, int y, float cx, float cy, float res, float clen, float k,
float *ttp, float4 z, int sampling)
{
float rx, ry, r;
float ttx, tty, tt;
float4 q;
rx = ((float)x - sampling*cx)/(res*sampling);
ry = ((float)y - sampling*cy)/(res*sampling);
r = sqrt(pow(rx, 2.0) + pow(ry, 2.0));
ttx = atan2(rx, clen);
tty = atan2(ry, clen);
tt = atan2(r, clen);
*ttp = tt;
q = (float4)(k*sin(ttx), k*sin(tty), k-k*cos(tt), 0.0);
return quat_rot(q, z);
}
float lattice_factor(float16 cell, float4 q, int4 ncells)
{
float f1, f2, f3;
float4 Udotq;
const int na = ncells.s0;
const int nb = ncells.s1;
const int nc = ncells.s2;
Udotq.x = cell.s0*q.x + cell.s1*q.y + cell.s2*q.z;
Udotq.y = cell.s3*q.x + cell.s4*q.y + cell.s5*q.z;
Udotq.z = cell.s6*q.x + cell.s7*q.y + cell.s8*q.z;
/* At exact Bragg condition, f1 = na */
f1 = sin(M_PI*(float)na*Udotq.x) / sin(M_PI*Udotq.x);
/* At exact Bragg condition, f2 = nb */
f2 = sin(M_PI*(float)nb*Udotq.y) / sin(M_PI*Udotq.y);
/* At exact Bragg condition, f3 = nc */
f3 = sin(M_PI*(float)nc*Udotq.z) / sin(M_PI*Udotq.z);
/* At exact Bragg condition, this will multiply the molecular
* part of the structure factor by the number of unit cells,
* as desired (more scattering from bigger crystal!) */
return f1 * f2 * f3;
}
float2 get_sfac(global float2 *sfacs, float16 cell, float4 q)
{
float hf, kf, lf;
int h, k, l;
int idx;
hf = cell.s0*q.x + cell.s1*q.y + cell.s2*q.z; /* h */
kf = cell.s3*q.x + cell.s4*q.y + cell.s5*q.z; /* k */
lf = cell.s6*q.x + cell.s7*q.y + cell.s8*q.z; /* l */
h = round(hf);
k = round(kf);
l = round(lf);
if ( (abs(h) > INDMAX) || (abs(k) > INDMAX) || (abs(l) > INDMAX) ) {
return 100.0;
}
if ( h < 0 ) h += IDIM;
if ( k < 0 ) k += IDIM;
if ( l < 0 ) l += IDIM;
if ( (h>=IDIM) || (k>=IDIM) || (l>=IDIM) ) return 100.0;
idx = h + (IDIM*k) + (IDIM*IDIM*l);
return sfacs[idx];
}
kernel void diffraction(global float2 *diff, global float *tt, float klow,
int w, float cx, float cy,
float res, float clen, float16 cell,
global float2 *sfacs, float4 z, int4 ncells,
int xmin, int ymin, int sampling, local float2 *tmp,
float kstep)
{
float ttv;
const int x = get_global_id(0) + (xmin*sampling);
const int y = get_global_id(1) + (ymin*sampling);
float f_lattice;
float2 f_molecule;
float4 q;
const int lx = get_local_id(0);
const int ly = get_local_id(1);
const int lb = get_local_id(2);
float k = klow + kstep * get_local_id(2);
const int ax = x / sampling;
const int ay = y / sampling;
/* Calculate value */
q = get_q(x, y, cx, cy, res, clen, k, &ttv, z, sampling);
f_lattice = lattice_factor(cell, q, ncells);
f_molecule = get_sfac(sfacs, cell, q);
/* Write the value to local memory */
tmp[lx+sampling*ly+sampling*sampling*lb] = f_molecule * f_lattice;
/* Memory fence */
barrier(CLK_LOCAL_MEM_FENCE);
/* Leader thread sums the values */
if ( lx + ly == 0 ) {
int i;
float2 sum = (0.0, 0.0);
for ( i=0; i<sampling*sampling*get_local_size(2); i++ )
sum += tmp[i];
diff[ax+w*ay] = sum / (sampling*sampling*get_local_size(2));
/* Leader thread also records the 2theta value.
* This should really be averaged across all pixels, but
* I strongly suspect this would be a waste of time. */
tt[ax+w*ay] = ttv;
}
}
|
