Neko 1.99.9
A portable framework for high-order spectral element flow simulations
Loading...
Searching...
No Matches
compressible_res.c
Go to the documentation of this file.
1/*
2 Copyright (c) 2025, 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>
42#include <device/device_config.h>
43#include <device/opencl/jit.h>
44#include <device/opencl/prgm_lib.h>
45#include <device/opencl/check.h>
46
47#include "compressible_res_kernel.cl.h"
48
49void compressible_res_part_visc_opencl(void *rhs_rho, void *rhs_m_x,
50 void *rhs_m_y, void *rhs_m_z, void *rhs_E,
51 void *visc_rho, void *visc_m_x, void *visc_m_y,
52 void *visc_m_z, void *visc_E,
53 void *Binv, void *h1, int *n) {
54 cl_int err;
55
56 if (compressible_res_program == NULL)
57 opencl_kernel_jit(compressible_res_kernel, (cl_program *) &compressible_res_program);
58
59 cl_kernel kernel = clCreateKernel(compressible_res_program,
60 "compressible_res_part_visc_kernel", &err);
61 CL_CHECK(err);
62
63 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &rhs_rho));
64 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &rhs_m_x));
65 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &rhs_m_y));
66 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &rhs_m_z));
67 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &rhs_E));
68 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &visc_rho));
69 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &visc_m_x));
70 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(cl_mem), (void *) &visc_m_y));
71 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &visc_m_z));
72 CL_CHECK(clSetKernelArg(kernel, 9, sizeof(cl_mem), (void *) &visc_E));
73 CL_CHECK(clSetKernelArg(kernel, 10, sizeof(cl_mem), (void *) &Binv));
74 CL_CHECK(clSetKernelArg(kernel, 11, sizeof(cl_mem), (void *) &h1));
75 CL_CHECK(clSetKernelArg(kernel, 12, sizeof(int), n));
76
77 const int nb = ((*n) + 256 - 1) / 256;
78 const size_t global_item_size = 256 * nb;
79 const size_t local_item_size = 256;
80
81 CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) glb_cmd_queue, kernel, 1,
82 NULL, &global_item_size, &local_item_size,
83 0, NULL, NULL));
84 CL_CHECK(clReleaseKernel(kernel));
85}
86
87void inviscid_res_part_mx_flux_opencl(void *f_x, void *f_y, void *f_z,
88 void *m_x, void *m_y, void *m_z,
89 void *rho_field, void *p, int *n) {
90 cl_int err;
91
92 if (compressible_res_program == NULL)
93 opencl_kernel_jit(compressible_res_kernel, (cl_program *) &compressible_res_program);
94
95 cl_kernel kernel = clCreateKernel(compressible_res_program,
96 "inviscid_res_part_mx_flux_kernel", &err);
97 CL_CHECK(err);
98
99 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &f_x));
100 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &f_y));
101 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &f_z));
102 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &m_x));
103 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &m_y));
104 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &m_z));
105 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &rho_field));
106 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(cl_mem), (void *) &p));
107 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(int), n));
108
109 const int nb = ((*n) + 256 - 1) / 256;
110 const size_t global_item_size = 256 * nb;
111 const size_t local_item_size = 256;
112
113 CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) glb_cmd_queue, kernel, 1,
114 NULL, &global_item_size, &local_item_size,
115 0, NULL, NULL));
116 CL_CHECK(clReleaseKernel(kernel));
117}
118
119void inviscid_res_part_my_flux_opencl(void *f_x, void *f_y, void *f_z,
120 void *m_x, void *m_y, void *m_z,
121 void *rho_field, void *p, int *n) {
122 cl_int err;
123
124 if (compressible_res_program == NULL)
125 opencl_kernel_jit(compressible_res_kernel, (cl_program *) &compressible_res_program);
126
127 cl_kernel kernel = clCreateKernel(compressible_res_program,
128 "inviscid_res_part_my_flux_kernel", &err);
129 CL_CHECK(err);
130
131 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &f_x));
132 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &f_y));
133 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &f_z));
134 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &m_x));
135 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &m_y));
136 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &m_z));
137 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &rho_field));
138 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(cl_mem), (void *) &p));
139 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(int), n));
140
141 const int nb = ((*n) + 256 - 1) / 256;
142 const size_t global_item_size = 256 * nb;
143 const size_t local_item_size = 256;
144
145 CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) glb_cmd_queue, kernel, 1,
146 NULL, &global_item_size, &local_item_size,
147 0, NULL, NULL));
148 CL_CHECK(clReleaseKernel(kernel));
149}
150
151void inviscid_res_part_mz_flux_opencl(void *f_x, void *f_y, void *f_z,
152 void *m_x, void *m_y, void *m_z,
153 void *rho_field, void *p, int *n) {
154 cl_int err;
155
156 if (compressible_res_program == NULL)
157 opencl_kernel_jit(compressible_res_kernel, (cl_program *) &compressible_res_program);
158
159 cl_kernel kernel = clCreateKernel(compressible_res_program,
160 "inviscid_res_part_mz_flux_kernel", &err);
161 CL_CHECK(err);
162
163 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &f_x));
164 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &f_y));
165 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &f_z));
166 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &m_x));
167 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &m_y));
168 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &m_z));
169 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &rho_field));
170 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(cl_mem), (void *) &p));
171 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(int), n));
172
173 const int nb = ((*n) + 256 - 1) / 256;
174 const size_t global_item_size = 256 * nb;
175 const size_t local_item_size = 256;
176
177 CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) glb_cmd_queue, kernel, 1,
178 NULL, &global_item_size, &local_item_size,
179 0, NULL, NULL));
180 CL_CHECK(clReleaseKernel(kernel));
181}
182
183void inviscid_res_part_E_flux_opencl(void *f_x, void *f_y, void *f_z,
184 void *m_x, void *m_y, void *m_z,
185 void *rho_field, void *p, void * E, int *n) {
186 cl_int err;
187
188 if (compressible_res_program == NULL)
189 opencl_kernel_jit(compressible_res_kernel, (cl_program *) &compressible_res_program);
190
191 cl_kernel kernel = clCreateKernel(compressible_res_program,
192 "inviscid_res_part_E_flux_kernel", &err);
193 CL_CHECK(err);
194
195 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &f_x));
196 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &f_y));
197 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &f_z));
198 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &m_x));
199 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &m_y));
200 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &m_z));
201 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &rho_field));
202 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(cl_mem), (void *) &p));
203 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &E));
204 CL_CHECK(clSetKernelArg(kernel, 9, sizeof(int), n));
205
206 const int nb = ((*n) + 256 - 1) / 256;
207 const size_t global_item_size = 256 * nb;
208 const size_t local_item_size = 256;
209
210 CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) glb_cmd_queue, kernel, 1,
211 NULL, &global_item_size, &local_item_size,
212 0, NULL, NULL));
213 CL_CHECK(clReleaseKernel(kernel));
214}
215
216void compressible_res_part_coef_mult_opencl(void *rhs_rho, void *rhs_m_x,
217 void *rhs_m_y, void *rhs_m_z,
218 void *rhs_E, void *mult, int *n) {
219 cl_int err;
220
221 if (compressible_res_program == NULL)
222 opencl_kernel_jit(compressible_res_kernel, (cl_program *) &compressible_res_program);
223
224 cl_kernel kernel = clCreateKernel(compressible_res_program,
225 "compressible_res_part_coef_mult_kernel", &err);
226 CL_CHECK(err);
227
228 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &rhs_rho));
229 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &rhs_m_x));
230 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &rhs_m_y));
231 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &rhs_m_z));
232 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &rhs_E));
233 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &mult));
234 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(int), n));
235
236 const int nb = ((*n) + 256 - 1) / 256;
237 const size_t global_item_size = 256 * nb;
238 const size_t local_item_size = 256;
239
240 CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) glb_cmd_queue, kernel, 1,
241 NULL, &global_item_size, &local_item_size,
242 0, NULL, NULL));
243 CL_CHECK(clReleaseKernel(kernel));
244}
245
246void compressible_res_part_rk_sum_opencl(void *rho, void *m_x, void *m_y, void *m_z,
247 void *E, void *k_rho_i, void *k_m_x_i,
248 void *k_m_y_i, void *k_m_z_i, void *k_E_i,
249 real *dt, real *c, int *n) {
250 cl_int err;
251
252 if (compressible_res_program == NULL)
253 opencl_kernel_jit(compressible_res_kernel, (cl_program *) &compressible_res_program);
254
255 cl_kernel kernel = clCreateKernel(compressible_res_program,
256 "compressible_res_part_rk_sum_kernel", &err);
257 CL_CHECK(err);
258
259 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &rho));
260 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &m_x));
261 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &m_y));
262 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &m_z));
263 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &E));
264 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &k_rho_i));
265 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &k_m_x_i));
266 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(cl_mem), (void *) &k_m_y_i));
267 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &k_m_z_i));
268 CL_CHECK(clSetKernelArg(kernel, 9, sizeof(cl_mem), (void *) &k_E_i));
269 CL_CHECK(clSetKernelArg(kernel, 10, sizeof(real), dt));
270 CL_CHECK(clSetKernelArg(kernel, 11, sizeof(real), c));
271 CL_CHECK(clSetKernelArg(kernel, 12, sizeof(int), n));
272
273 const int nb = ((*n) + 256 - 1) / 256;
274 const size_t global_item_size = 256 * nb;
275 const size_t local_item_size = 256;
276
277 CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) glb_cmd_queue, kernel, 1,
278 NULL, &global_item_size, &local_item_size,
279 0, NULL, NULL));
280 CL_CHECK(clReleaseKernel(kernel));
281}
compressible_res_part_visc_opencl
void compressible_res_part_visc_opencl(void *rhs_rho, void *rhs_m_x, void *rhs_m_y, void *rhs_m_z, void *rhs_E, void *visc_rho, void *visc_m_x, void *visc_m_y, void *visc_m_z, void *visc_E, void *Binv, void *h1, int *n)
Definition compressible_res.c:49
inviscid_res_part_E_flux_opencl
void inviscid_res_part_E_flux_opencl(void *f_x, void *f_y, void *f_z, void *m_x, void *m_y, void *m_z, void *rho_field, void *p, void *E, int *n)
Definition compressible_res.c:183
inviscid_res_part_mz_flux_opencl
void inviscid_res_part_mz_flux_opencl(void *f_x, void *f_y, void *f_z, void *m_x, void *m_y, void *m_z, void *rho_field, void *p, int *n)
Definition compressible_res.c:151
inviscid_res_part_my_flux_opencl
void inviscid_res_part_my_flux_opencl(void *f_x, void *f_y, void *f_z, void *m_x, void *m_y, void *m_z, void *rho_field, void *p, int *n)
Definition compressible_res.c:119
compressible_res_part_rk_sum_opencl
void compressible_res_part_rk_sum_opencl(void *rho, void *m_x, void *m_y, void *m_z, void *E, void *k_rho_i, void *k_m_x_i, void *k_m_y_i, void *k_m_z_i, void *k_E_i, real *dt, real *c, int *n)
Definition compressible_res.c:246
compressible_res_part_coef_mult_opencl
void compressible_res_part_coef_mult_opencl(void *rhs_rho, void *rhs_m_x, void *rhs_m_y, void *rhs_m_z, void *rhs_E, void *mult, int *n)
Definition compressible_res.c:216
inviscid_res_part_mx_flux_opencl
void inviscid_res_part_mx_flux_opencl(void *f_x, void *f_y, void *f_z, void *m_x, void *m_y, void *m_z, void *rho_field, void *p, int *n)
Definition compressible_res.c:87
ale_add_kinematics_kernel
__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)
Definition ale_kinematics_kernel.h:52
h1
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ h1
Definition ax_helm_full_kernel.h:52
dt
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dt
Definition cdtp_kernel.h:113
device_config.h
real
double real
Definition device_config.h:12
opencl_kernel_jit
void opencl_kernel_jit(const char *kernel, cl_program *program)
Definition jit.c:50
CL_CHECK
#define CL_CHECK(err)
Definition check.h:12
prgm_lib.h
compressible_res_program
void * compressible_res_program