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
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
|
#include <stdio.h>
#include <time.h>
#include <sys/types.h>
#include <sys/time.h>
#include "../gpu.h"
#include "../gpu-complex.h"
#ifndef GPU_USE_COMPLEX
# define BLOCK_SIZEX 32
# define BLOCK_SIZEY 16
# define BLOCK_MAX 32
# define GRID_MAX 65535
# define BLOCK_TOTALMAX 512
# define XDIV 4
# define Tfunc launch_func
typedef void (*launch_func)(const double *, const int *,
double *, const int *, const int *, int,
gpuStream_t);
typedef void (*launch_funcz)(const gpuDoubleComplex *, const int *,
gpuDoubleComplex *, const int *, const int *, int,
gpuStream_t);
#else
# undef Tfunc
# define Tfunc launch_funcz
#endif
/*
* GPU kernel to copy a smaller array into a given position in a
* larger one.
*/
__global__ void Zgpu(bmgs_paste_kernel)(
const double* a, const int3 c_sizea, double* b, const int3 c_sizeb,
int blocks, int xdiv)
{
int xx = gridDim.x / xdiv;
int yy = gridDim.y / blocks;
int blocksi = blockIdx.y / yy;
int i1 = (blockIdx.y - blocksi * yy) * blockDim.y + threadIdx.y;
int xind = blockIdx.x / xx;
int i2 = (blockIdx.x - xind * xx) * blockDim.x + threadIdx.x;
b += i2 + (i1 + (xind + blocksi * c_sizeb.x) * c_sizeb.y) * c_sizeb.z;
a += i2 + (i1 + (xind + blocksi * c_sizea.x) * c_sizea.y) * c_sizea.z;
while (xind < c_sizea.x) {
if ((i2 < c_sizea.z) && (i1 < c_sizea.y)) {
b[0] = a[0];
}
b += xdiv * c_sizeb.y * c_sizeb.z;
a += xdiv * c_sizea.y * c_sizea.z;
xind += xdiv;
}
}
/*
* GPU kernel to copy a smaller array into a given position in a
* larger one and set all other elements to 0.
*/
__global__ void Zgpu(bmgs_paste_zero_kernel)(
const Tgpu* a, const int3 c_sizea, Tgpu* b, const int3 c_sizeb,
const int3 c_startb, const int3 c_blocks_bc, int blocks)
{
int xx = gridDim.x / XDIV;
int yy = gridDim.y / blocks;
int blocksi = blockIdx.y / yy;
int i1bl = blockIdx.y - blocksi * yy;
int i1tid = threadIdx.y;
int i1 = i1bl * BLOCK_SIZEY + i1tid;
int xind = blockIdx.x / xx;
int i2bl = blockIdx.x - xind * xx;
int i2tid = threadIdx.x;
int i2 = i2bl * BLOCK_SIZEX + i2tid;
int xlen = (c_sizea.x + XDIV - 1) / XDIV;
int xstart = xind * xlen;
int xend = MIN(xstart + xlen, c_sizea.x);
b += c_sizeb.x * c_sizeb.y * c_sizeb.z * blocksi;
a += c_sizea.x * c_sizea.y * c_sizea.z * blocksi;
// zero x = 0 .. startb.x
if (xind==0) {
Tgpu *bb = b + i2 + i1 * c_sizeb.z;
#pragma unroll 3
for (int i0=0; i0 < c_startb.x; i0++) {
if ((i2 < c_sizeb.z) && (i1 < c_sizeb.y)) {
bb[0] = MAKED(0);
}
bb += c_sizeb.y * c_sizeb.z;
}
}
// zero x = startb.x+sizea.x .. <end>
if (xind == XDIV - 1) {
Tgpu *bb = b + (c_startb.x + c_sizea.x) * c_sizeb.y * c_sizeb.z
+ i2 + i1 * c_sizeb.z;
#pragma unroll 3
for (int i0 = c_startb.x + c_sizea.x; i0 < c_sizeb.x; i0++) {
if ((i2 < c_sizeb.z) && (i1 < c_sizeb.y)) {
bb[0] = MAKED(0);
}
bb += c_sizeb.y * c_sizeb.z;
}
}
int i1blbc = gridDim.y / blocks - i1bl - 1;
int i2blbc = gridDim.x / XDIV - i2bl - 1;
if (i1blbc<c_blocks_bc.y || i2blbc<c_blocks_bc.z) {
int i1bc = i1blbc * BLOCK_SIZEY + i1tid;
int i2bc = i2blbc * BLOCK_SIZEX + i2tid;
b += (c_startb.x + xstart) * c_sizeb.y * c_sizeb.z;
for (int i0=xstart; i0 < xend; i0++) {
// zero y = 0 .. startb.y
if ((i1bc < c_startb.y) && (i2 < c_sizeb.z)) {
b[i2 + i1bc * c_sizeb.z] = MAKED(0);
}
// zero y = startb.y+sizea.y .. <end>
if ((i1bc + c_sizea.y + c_startb.y < c_sizeb.y)
&& (i2 < c_sizeb.z)) {
b[i2 + i1bc * c_sizeb.z
+ (c_sizea.y + c_startb.y) * c_sizeb.z] = MAKED(0);
}
// zero z = 0 .. startb.z
if ((i2bc < c_startb.z) && (i1 < c_sizeb.y)) {
b[i2bc + i1 * c_sizeb.z] = MAKED(0);
}
// zero z = startb.z+sizea.z .. <end>
if ((i2bc + c_sizea.z + c_startb.z < c_sizeb.z)
&& (i1 < c_sizeb.y)) {
b[i2bc + i1 * c_sizeb.z + c_sizea.z + c_startb.z] = MAKED(0);
}
b += c_sizeb.y * c_sizeb.z;
}
} else {
b += c_startb.z + (c_startb.y + c_startb.x * c_sizeb.y) * c_sizeb.z;
b += i2 + i1 * c_sizeb.z + xstart * c_sizeb.y * c_sizeb.z;
a += i2 + i1 * c_sizea.z + xstart * c_sizea.y * c_sizea.z;
for (int i0=xstart; i0 < xend; i0++) {
if ((i2 < c_sizea.z) && (i1 < c_sizea.y)) {
b[0] = a[0];
}
b += c_sizeb.y * c_sizeb.z;
a += c_sizea.y * c_sizea.z;
}
}
}
/*
* Copy a smaller array into a given position in a larger one.
*
* For example:
* . . . .
* a = 1 2 -> b = . 1 2 .
* 3 4 . 3 4 .
* . . . .
*/
extern "C"
void Zgpu(bmgs_paste_gpu)(const Tgpu* a, const int sizea[3],
Tgpu* b, const int sizeb[3],
const int startb[3], int blocks,
gpuStream_t stream)
{
if (!(sizea[0] && sizea[1] && sizea[2]))
return;
int3 hc_sizea, hc_sizeb;
hc_sizea.x = sizea[0];
hc_sizea.y = sizea[1];
hc_sizea.z = sizea[2] * sizeof(Tgpu) / sizeof(double);
hc_sizeb.x = sizeb[0];
hc_sizeb.y = sizeb[1];
hc_sizeb.z = sizeb[2] * sizeof(Tgpu) / sizeof(double);
BLOCK_GRID(hc_sizea);
b += startb[2] + (startb[1] + startb[0] * sizeb[1]) * sizeb[2];
gpuLaunchKernel(
Zgpu(bmgs_paste_kernel), dimGrid, dimBlock, 0, stream,
(double*) a, hc_sizea, (double*) b, hc_sizeb, blocks, xdiv);
gpuCheckLastError();
}
/*
* Copy a smaller array into a given position in a larger one and
* set all other elements to 0.
*/
extern "C"
void Zgpu(bmgs_paste_zero_gpu)(const Tgpu* a, const int sizea[3],
Tgpu* b, const int sizeb[3],
const int startb[3], int blocks,
gpuStream_t stream)
{
if (!(sizea[0] && sizea[1] && sizea[2]))
return;
int3 bc_blocks;
int3 hc_sizea, hc_sizeb, hc_startb;
hc_sizea.x = sizea[0];
hc_sizea.y = sizea[1];
hc_sizea.z = sizea[2];
hc_sizeb.x = sizeb[0];
hc_sizeb.y = sizeb[1];
hc_sizeb.z = sizeb[2];
hc_startb.x = startb[0];
hc_startb.y = startb[1];
hc_startb.z = startb[2];
bc_blocks.y = hc_sizeb.y - hc_sizea.y > 0
? MAX((hc_sizeb.y - hc_sizea.y + BLOCK_SIZEY - 1)
/ BLOCK_SIZEY, 1)
: 0;
bc_blocks.z = hc_sizeb.z - hc_sizea.z > 0
? MAX((hc_sizeb.z - hc_sizea.z + BLOCK_SIZEX - 1)
/ BLOCK_SIZEX, 1)
: 0;
int gridy = blocks * ((sizeb[1] + BLOCK_SIZEY - 1) / BLOCK_SIZEY
+ bc_blocks.y);
int gridx = XDIV * ((sizeb[2] + BLOCK_SIZEX - 1) / BLOCK_SIZEX
+ bc_blocks.z);
dim3 dimBlock(BLOCK_SIZEX, BLOCK_SIZEY);
dim3 dimGrid(gridx, gridy);
gpuLaunchKernel(
Zgpu(bmgs_paste_zero_kernel), dimGrid, dimBlock, 0, stream,
(Tgpu*) a, hc_sizea, (Tgpu*) b, hc_sizeb, hc_startb,
bc_blocks, blocks);
gpuCheckLastError();
}
#ifndef GPU_USE_COMPLEX
#define GPU_USE_COMPLEX
#include "paste.cpp"
#endif
|
