-
Notifications
You must be signed in to change notification settings - Fork 1
/
gneb_sumimage_wrapper.go
282 lines (239 loc) · 6.73 KB
/
gneb_sumimage_wrapper.go
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
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
package cuda
/*
THIS FILE IS AUTO-GENERATED BY CUDA2GO.
EDITING IS FUTILE.
*/
import (
"github.com/kuchkin/mumax3-gneb/cuda/cu"
"github.com/kuchkin/mumax3-gneb/timer"
"sync"
"unsafe"
)
// CUDA handle for sumimage kernel
var sumimage_code cu.Function
// Stores the arguments for sumimage kernel invocation
type sumimage_args_t struct {
arg_src unsafe.Pointer
arg_dst unsafe.Pointer
arg_initVal float32
arg_n int
arg_image int
arg_noi int
arg_Nz int
argptr [7]unsafe.Pointer
sync.Mutex
}
// Stores the arguments for sumimage kernel invocation
var sumimage_args sumimage_args_t
func init() {
// CUDA driver kernel call wants pointers to arguments, set them up once.
sumimage_args.argptr[0] = unsafe.Pointer(&sumimage_args.arg_src)
sumimage_args.argptr[1] = unsafe.Pointer(&sumimage_args.arg_dst)
sumimage_args.argptr[2] = unsafe.Pointer(&sumimage_args.arg_initVal)
sumimage_args.argptr[3] = unsafe.Pointer(&sumimage_args.arg_n)
sumimage_args.argptr[4] = unsafe.Pointer(&sumimage_args.arg_image)
sumimage_args.argptr[5] = unsafe.Pointer(&sumimage_args.arg_noi)
sumimage_args.argptr[6] = unsafe.Pointer(&sumimage_args.arg_Nz)
}
// Wrapper for sumimage CUDA kernel, asynchronous.
func k_sumimage_async(src unsafe.Pointer, dst unsafe.Pointer, initVal float32, n int, image int, noi int, Nz int, cfg *config) {
if Synchronous { // debug
Sync()
timer.Start("sumimage")
}
sumimage_args.Lock()
defer sumimage_args.Unlock()
if sumimage_code == 0 {
sumimage_code = fatbinLoad(sumimage_map, "sumimage")
}
sumimage_args.arg_src = src
sumimage_args.arg_dst = dst
sumimage_args.arg_initVal = initVal
sumimage_args.arg_n = n
sumimage_args.arg_image = image
sumimage_args.arg_noi = noi
sumimage_args.arg_Nz = Nz
args := sumimage_args.argptr[:]
cu.LaunchKernel(sumimage_code, cfg.Grid.X, cfg.Grid.Y, cfg.Grid.Z, cfg.Block.X, cfg.Block.Y, cfg.Block.Z, 0, stream0, args)
if Synchronous { // debug
Sync()
timer.Stop("sumimage")
}
}
// maps compute capability on PTX code for sumimage kernel.
var sumimage_map = map[int]string{0: "",
50: sumimage_ptx_50}
// sumimage PTX code for various compute capabilities.
const (
sumimage_ptx_50 = `
.version 7.5
.target sm_50
.address_size 64
// .globl sumimage
.visible .entry sumimage(
.param .u64 sumimage_param_0,
.param .u64 sumimage_param_1,
.param .f32 sumimage_param_2,
.param .u32 sumimage_param_3,
.param .u32 sumimage_param_4,
.param .u32 sumimage_param_5,
.param .u32 sumimage_param_6
)
{
.reg .pred %p<16>;
.reg .f32 %f<56>;
.reg .b32 %r<52>;
.reg .b64 %rd<19>;
// demoted variable
.shared .align 4 .b8 _ZZ8sumimageE5sdata[2048];
ld.param.u64 %rd7, [sumimage_param_0];
ld.param.u64 %rd6, [sumimage_param_1];
ld.param.f32 %f51, [sumimage_param_2];
ld.param.u32 %r22, [sumimage_param_3];
ld.param.u32 %r23, [sumimage_param_4];
ld.param.u32 %r24, [sumimage_param_5];
ld.param.u32 %r25, [sumimage_param_6];
cvta.to.global.u64 %rd1, %rd7;
mov.u32 %r51, %ntid.x;
mov.u32 %r26, %ctaid.x;
mov.u32 %r2, %tid.x;
mad.lo.s32 %r49, %r26, %r51, %r2;
mov.u32 %r27, %nctaid.x;
mul.lo.s32 %r4, %r27, %r51;
setp.ge.s32 %p1, %r49, %r22;
@%p1 bra $L__BB0_17;
div.s32 %r5, %r22, %r25;
div.s32 %r6, %r25, %r24;
add.s32 %r28, %r49, %r4;
not.b32 %r29, %r28;
add.s32 %r30, %r4, %r22;
add.s32 %r31, %r30, %r29;
div.u32 %r7, %r31, %r4;
add.s32 %r32, %r7, 1;
and.b32 %r48, %r32, 3;
setp.eq.s32 %p2, %r48, 0;
@%p2 bra $L__BB0_6;
mul.wide.s32 %rd8, %r49, 4;
add.s64 %rd18, %rd1, %rd8;
mul.wide.s32 %rd3, %r4, 4;
$L__BB0_3:
.pragma "nounroll";
div.s32 %r33, %r49, %r5;
div.s32 %r34, %r33, %r6;
setp.ne.s32 %p3, %r34, %r23;
@%p3 bra $L__BB0_5;
ld.global.nc.f32 %f18, [%rd18];
add.f32 %f51, %f51, %f18;
$L__BB0_5:
add.s32 %r49, %r49, %r4;
add.s64 %rd18, %rd18, %rd3;
add.s32 %r48, %r48, -1;
setp.ne.s32 %p4, %r48, 0;
@%p4 bra $L__BB0_3;
$L__BB0_6:
setp.lt.u32 %p5, %r7, 3;
@%p5 bra $L__BB0_17;
$L__BB0_8:
div.s32 %r35, %r49, %r5;
div.s32 %r36, %r35, %r6;
setp.ne.s32 %p6, %r36, %r23;
@%p6 bra $L__BB0_10;
mul.wide.s32 %rd9, %r49, 4;
add.s64 %rd10, %rd1, %rd9;
ld.global.nc.f32 %f19, [%rd10];
add.f32 %f51, %f51, %f19;
$L__BB0_10:
add.s32 %r15, %r49, %r4;
div.s32 %r37, %r15, %r5;
div.s32 %r38, %r37, %r6;
setp.ne.s32 %p7, %r38, %r23;
@%p7 bra $L__BB0_12;
mul.wide.s32 %rd11, %r15, 4;
add.s64 %rd12, %rd1, %rd11;
ld.global.nc.f32 %f20, [%rd12];
add.f32 %f51, %f51, %f20;
$L__BB0_12:
add.s32 %r16, %r15, %r4;
div.s32 %r39, %r16, %r5;
div.s32 %r40, %r39, %r6;
setp.ne.s32 %p8, %r40, %r23;
@%p8 bra $L__BB0_14;
mul.wide.s32 %rd13, %r16, 4;
add.s64 %rd14, %rd1, %rd13;
ld.global.nc.f32 %f21, [%rd14];
add.f32 %f51, %f51, %f21;
$L__BB0_14:
add.s32 %r17, %r16, %r4;
div.s32 %r41, %r17, %r5;
div.s32 %r42, %r41, %r6;
setp.ne.s32 %p9, %r42, %r23;
@%p9 bra $L__BB0_16;
mul.wide.s32 %rd15, %r17, 4;
add.s64 %rd16, %rd1, %rd15;
ld.global.nc.f32 %f22, [%rd16];
add.f32 %f51, %f51, %f22;
$L__BB0_16:
add.s32 %r49, %r17, %r4;
setp.lt.s32 %p10, %r49, %r22;
@%p10 bra $L__BB0_8;
$L__BB0_17:
shl.b32 %r43, %r2, 2;
mov.u32 %r44, _ZZ8sumimageE5sdata;
add.s32 %r19, %r44, %r43;
st.shared.f32 [%r19], %f51;
bar.sync 0;
setp.lt.u32 %p11, %r51, 66;
@%p11 bra $L__BB0_22;
$L__BB0_19:
shr.u32 %r21, %r51, 1;
setp.ge.u32 %p12, %r2, %r21;
@%p12 bra $L__BB0_21;
ld.shared.f32 %f23, [%r19];
shl.b32 %r45, %r21, 2;
add.s32 %r46, %r19, %r45;
ld.shared.f32 %f24, [%r46];
add.f32 %f25, %f23, %f24;
st.shared.f32 [%r19], %f25;
$L__BB0_21:
bar.sync 0;
setp.gt.u32 %p13, %r51, 131;
mov.u32 %r51, %r21;
@%p13 bra $L__BB0_19;
$L__BB0_22:
setp.gt.s32 %p14, %r2, 31;
@%p14 bra $L__BB0_24;
ld.volatile.shared.f32 %f26, [%r19];
ld.volatile.shared.f32 %f27, [%r19+128];
add.f32 %f28, %f26, %f27;
st.volatile.shared.f32 [%r19], %f28;
ld.volatile.shared.f32 %f29, [%r19+64];
ld.volatile.shared.f32 %f30, [%r19];
add.f32 %f31, %f30, %f29;
st.volatile.shared.f32 [%r19], %f31;
ld.volatile.shared.f32 %f32, [%r19+32];
ld.volatile.shared.f32 %f33, [%r19];
add.f32 %f34, %f33, %f32;
st.volatile.shared.f32 [%r19], %f34;
ld.volatile.shared.f32 %f35, [%r19+16];
ld.volatile.shared.f32 %f36, [%r19];
add.f32 %f37, %f36, %f35;
st.volatile.shared.f32 [%r19], %f37;
ld.volatile.shared.f32 %f38, [%r19+8];
ld.volatile.shared.f32 %f39, [%r19];
add.f32 %f40, %f39, %f38;
st.volatile.shared.f32 [%r19], %f40;
ld.volatile.shared.f32 %f41, [%r19+4];
ld.volatile.shared.f32 %f42, [%r19];
add.f32 %f43, %f42, %f41;
st.volatile.shared.f32 [%r19], %f43;
$L__BB0_24:
setp.ne.s32 %p15, %r2, 0;
@%p15 bra $L__BB0_26;
ld.shared.f32 %f44, [_ZZ8sumimageE5sdata];
cvta.to.global.u64 %rd17, %rd6;
atom.global.add.f32 %f45, [%rd17], %f44;
$L__BB0_26:
ret;
}
`
)