Neko 1.99.7
A portable framework for high-order spectral element flow simulations
Loading...
Searching...
No Matches
rhs_maker.c
Go to the documentation of this file.
1/*
2 Copyright (c) 2022, The Neko Authors
3 All rights reserved.
4
5 Redistribution and use in source and binary forms, with or without
6 modification, are permitted provided that the following conditions
7 are met:
8
9 * Redistributions of source code must retain the above copyright
10 notice, this list of conditions and the following disclaimer.
11
12 * Redistributions in binary form must reproduce the above
13 copyright notice, this list of conditions and the following
14 disclaimer in the documentation and/or other materials provided
15 with the distribution.
16
17 * Neither the name of the authors nor the names of its
18 contributors may be used to endorse or promote products derived
19 from this software without specific prior written permission.
20
21 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
29 CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
31 ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 POSSIBILITY OF SUCH DAMAGE.
33*/
34
35#ifdef __APPLE__
36#include <OpenCL/cl.h>
37#else
38#include <CL/cl.h>
39#endif
40
41#include <stdio.h>
43#include <device/opencl/jit.h>
45#include <device/opencl/check.h>
46
47#include "rhs_maker_kernel.cl.h"
48
49void rhs_maker_sumab_opencl(void *u, void *v, void *w,
50 void *uu, void *vv, void *ww,
51 void *ulag1, void *ulag2, void *vlag1,
52 void *vlag2, void *wlag1, void *wlag2,
53 real *ext1, real *ext2, real *ext3, int *nab, int *n) {
54 cl_int err;
55
58
61
62 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &u));
63 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &v));
64 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &w));
65 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &uu));
66 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &vv));
67 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &ww));
68 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &ulag1));
69 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(cl_mem), (void *) &ulag2));
70 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &vlag1));
71 CL_CHECK(clSetKernelArg(kernel, 9, sizeof(cl_mem), (void *) &vlag2));
72 CL_CHECK(clSetKernelArg(kernel, 10, sizeof(cl_mem), (void *) &wlag1));
73 CL_CHECK(clSetKernelArg(kernel, 11, sizeof(cl_mem), (void *) &wlag2));
74 CL_CHECK(clSetKernelArg(kernel, 12, sizeof(real), ext1));
75 CL_CHECK(clSetKernelArg(kernel, 13, sizeof(real), ext2));
76 CL_CHECK(clSetKernelArg(kernel, 14, sizeof(real), ext3));
77 CL_CHECK(clSetKernelArg(kernel, 15, sizeof(int), nab));
78 CL_CHECK(clSetKernelArg(kernel, 16, sizeof(int), n));
79
80 const int nb = ((*n) + 256 - 1) / 256;
81 const size_t global_item_size = 256 * nb;
82 const size_t local_item_size = 256;
83
86 0, NULL, NULL));
88}
89
90void rhs_maker_ext_opencl(void *abx1, void *aby1, void *abz1,
91 void *abx2, void *aby2, void *abz2,
92 void *bfx, void *bfy, void *bfz,
93 real *rho, real *ext1, real *ext2, real *ext3, int *n) {
94 cl_int err;
95
98
100 CL_CHECK(err);
101
102 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &abx1));
103 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &aby1));
104 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &abz1));
105 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &abx2));
106 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &aby2));
107 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &abz2));
108 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &bfx));
109 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(cl_mem), (void *) &bfy));
110 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &bfz));
111 CL_CHECK(clSetKernelArg(kernel, 9, sizeof(real), rho));
112 CL_CHECK(clSetKernelArg(kernel, 10, sizeof(real), ext1));
113 CL_CHECK(clSetKernelArg(kernel, 11, sizeof(real), ext2));
114 CL_CHECK(clSetKernelArg(kernel, 12, sizeof(real), ext3));
115 CL_CHECK(clSetKernelArg(kernel, 13, sizeof(int), n));
116
117 const int nb = ((*n) + 256 - 1) / 256;
118 const size_t global_item_size = 256 * nb;
119 const size_t local_item_size = 256;
120
123 0, NULL, NULL));
124
126}
127
129 real *ext1, real *ext2, real *ext3, int *n) {
130 cl_int err;
131
132 if (rhs_maker_program == NULL)
134
135 cl_kernel kernel = clCreateKernel(rhs_maker_program, "scalar_makeext_kernel", &err);
136 CL_CHECK(err);
137
138 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &fs_lag));
139 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &fs_laglag));
140 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &fs));
141 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(real), ext1));
142 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(real), ext2));
143 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(real), ext3));
144 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(int), n));
145
146 const int nb = ((*n) + 256 - 1) / 256;
147 const size_t global_item_size = 256 * nb;
148 const size_t local_item_size = 256;
149
152 0, NULL, NULL));
153
155}
156
157void rhs_maker_bdf_opencl(void *ulag1, void *ulag2, void *vlag1,
158 void *vlag2, void *wlag1, void *wlag2,
159 void *bfx, void *bfy, void *bfz,
160 void *u, void *v, void *w, void *B,
161 real *rho, real *dt, real *bd2,
162 real *bd3, real *bd4, int *nbd, int *n) {
163 cl_int err;
164
165 if (rhs_maker_program == NULL)
167
168 cl_kernel kernel = clCreateKernel(rhs_maker_program, "makebdf_kernel", &err);
169 CL_CHECK(err);
170
171 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &ulag1));
172 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &ulag2));
173 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &vlag1));
174 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &vlag2));
175 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &wlag1));
176 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &wlag2));
177 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &bfx));
178 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(cl_mem), (void *) &bfy));
179 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &bfz));
180 CL_CHECK(clSetKernelArg(kernel, 9, sizeof(cl_mem), (void *) &u));
181 CL_CHECK(clSetKernelArg(kernel, 10, sizeof(cl_mem), (void *) &v));
182 CL_CHECK(clSetKernelArg(kernel, 11, sizeof(cl_mem), (void *) &w));
183 CL_CHECK(clSetKernelArg(kernel, 12, sizeof(cl_mem), (void *) &B));
184 CL_CHECK(clSetKernelArg(kernel, 13, sizeof(real), rho));
185 CL_CHECK(clSetKernelArg(kernel, 14, sizeof(real), dt));
186 CL_CHECK(clSetKernelArg(kernel, 15, sizeof(real), bd2));
187 CL_CHECK(clSetKernelArg(kernel, 16, sizeof(real), bd3));
188 CL_CHECK(clSetKernelArg(kernel, 17, sizeof(real), bd4));
189 CL_CHECK(clSetKernelArg(kernel, 18, sizeof(int), nbd));
190 CL_CHECK(clSetKernelArg(kernel, 19, sizeof(int), n));
191
192 const int nb = ((*n) + 256 - 1) / 256;
193 const size_t global_item_size = 256 * nb;
194 const size_t local_item_size = 256;
195
198 0, NULL, NULL));
199
201}
202
203void scalar_rhs_maker_bdf_opencl(void *s_lag, void *s_laglag, void *fs,
204 void *s, void *B, void *rho_cp, real *dt,
205 real *bd2, real *bd3, real *bd4,
206 int *nbd, int *n) {
207 cl_int err;
208
209 if (rhs_maker_program == NULL)
211
212 cl_kernel kernel = clCreateKernel(rhs_maker_program, "scalar_makebdf_kernel", &err);
213 CL_CHECK(err);
214
215 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &s_lag));
216 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &s_laglag));
217 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &fs));
218 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &s));
219 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &B));
220 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &rho_cp));
221 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(real), dt));
222 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(real), bd2));
223 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(real), bd3));
224 CL_CHECK(clSetKernelArg(kernel, 9, sizeof(real), bd4));
225 CL_CHECK(clSetKernelArg(kernel, 10, sizeof(int), nbd));
226 CL_CHECK(clSetKernelArg(kernel, 11, sizeof(int), n));
227
228 const int nb = ((*n) + 256 - 1) / 256;
229 const size_t global_item_size = 256 * nb;
230 const size_t local_item_size = 256;
231
234 0, NULL, NULL));
235
237}
238
239void rhs_maker_oifs_opencl(void *phi_x, void *phi_y, void *phi_z,
240 void *bf_x, void *bf_y, void *bf_z,
241 real *rho, real *dt, int *n) {
242 cl_int err;
243
244 if (rhs_maker_program == NULL)
246
247 cl_kernel kernel = clCreateKernel(rhs_maker_program, "makeoifs_kernel", &err);
248 CL_CHECK(err);
249
250 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &phi_x));
251 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &phi_y));
252 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &phi_z));
253 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &bf_x));
254 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &bf_y));
255 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &bf_z));
256 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(real), rho));
257 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(real), dt));
258 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(int), n));
259
260 const int nb = ((*n) + 256 - 1) / 256;
261 const size_t global_item_size = 256 * nb;
262 const size_t local_item_size = 256;
263
266 0, NULL, NULL));
267
269}
270
272 void *rho_cp, real *dt, int *n) {
273 cl_int err;
274
275 if (rhs_maker_program == NULL)
277
278 cl_kernel kernel = clCreateKernel(rhs_maker_program, "scalar_makeoifs_kernel", &err);
279 CL_CHECK(err);
280
281 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &phi_s));
282 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &bf_s));
283 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &rho_cp));
284 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(real), dt));
285 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(int), n));
286
287 const int nb = ((*n) + 256 - 1) / 256;
288 const size_t global_item_size = 256 * nb;
289 const size_t local_item_size = 256;
290
293 0, NULL, NULL));
294
296}
__global__ void ale_add_kinematics_kernel(const int n, T *__restrict__ wx, T *__restrict__ wy, T *__restrict__ wz, const T *__restrict__ x_ref, const T *__restrict__ y_ref, const T *__restrict__ z_ref, const T *__restrict__ phi, const T *__restrict__ x, const T *__restrict__ y, const T *__restrict__ z, const kinematics_params_t kin_params)
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ w
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ u
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ const T *__restrict__ v
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dt
double real
void opencl_kernel_jit(const char *kernel, cl_program *program)
Definition jit.c:50
#define CL_CHECK(err)
Definition check.h:12
void * rhs_maker_program
void rhs_maker_sumab_opencl(void *u, void *v, void *w, void *uu, void *vv, void *ww, void *ulag1, void *ulag2, void *vlag1, void *vlag2, void *wlag1, void *wlag2, real *ext1, real *ext2, real *ext3, int *nab, int *n)
Definition rhs_maker.c:49
void scalar_rhs_maker_bdf_opencl(void *s_lag, void *s_laglag, void *fs, void *s, void *B, void *rho_cp, real *dt, real *bd2, real *bd3, real *bd4, int *nbd, int *n)
Definition rhs_maker.c:203
void scalar_rhs_maker_ext_opencl(void *fs_lag, void *fs_laglag, void *fs, real *ext1, real *ext2, real *ext3, int *n)
Definition rhs_maker.c:128
void rhs_maker_ext_opencl(void *abx1, void *aby1, void *abz1, void *abx2, void *aby2, void *abz2, void *bfx, void *bfy, void *bfz, real *rho, real *ext1, real *ext2, real *ext3, int *n)
Definition rhs_maker.c:90
void rhs_maker_oifs_opencl(void *phi_x, void *phi_y, void *phi_z, void *bf_x, void *bf_y, void *bf_z, real *rho, real *dt, int *n)
Definition rhs_maker.c:239
void scalar_rhs_maker_oifs_opencl(void *phi_s, void *bf_s, void *rho_cp, real *dt, int *n)
Definition rhs_maker.c:271
void rhs_maker_bdf_opencl(void *ulag1, void *ulag2, void *vlag1, void *vlag2, void *wlag1, void *wlag2, void *bfx, void *bfy, void *bfz, void *u, void *v, void *w, void *B, real *rho, real *dt, real *bd2, real *bd3, real *bd4, int *nbd, int *n)
Definition rhs_maker.c:157