Neko 1.99.5
A portable framework for high-order spectral element flow simulations
Loading...
Searching...
No Matches
gs.c
Go to the documentation of this file.
1/*
2 Copyright (c) 2021-2025, 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 <limits.h>
42#include <stdio.h>
43#include <device/device_config.h>
44#include <device/opencl/jit.h>
45#include <device/opencl/prgm_lib.h>
46#include <device/opencl/check.h>
47
48#include "gs_kernels.cl.h"
49
50#define GS_OP_ADD 1
51#define GS_OP_MUL 2
52#define GS_OP_MIN 3
53#define GS_OP_MAX 4
54
58void opencl_gather_kernel(void *v, int *m, int *o, void *dg,
59 void *u, int *n, void *gd, int *nb,
60 void *b, void *bo, int *op, void *cmd_queue) {
61 cl_int err;
62
63 if (gs_program == NULL)
64 opencl_kernel_jit(gs_kernels, (cl_program *) &gs_program);
65
66 const int nblks = ((*m) + 256 - 1) / 256;
67 const size_t global_item_size = 256 * nblks;
68 const size_t local_item_size = 256;
69
70
71 switch (*op) {
72 case GS_OP_ADD:
73 {
74 cl_kernel kernel = clCreateKernel(gs_program,
75 "gather_kernel_add", &err);
76 CL_CHECK(err);
77
78 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &v));
79 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(int), m));
80 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(int), o));
81 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &dg));
82 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &u));
83 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(int), n));
84 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &gd));
85 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(int), nb));
86 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &b));
87 CL_CHECK(clSetKernelArg(kernel, 9, sizeof(cl_mem), (void *) &bo));
88
89 CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) cmd_queue, kernel,
90 1, NULL, &global_item_size,
91 &local_item_size, 0, NULL, NULL));
92 CL_CHECK(clReleaseKernel(kernel));
93 }
94 break;
95 case GS_OP_MUL:
96 {
97 cl_kernel kernel = clCreateKernel(gs_program,
98 "gather_kernel_mul", &err);
99 CL_CHECK(err);
100
101 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &v));
102 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(int), m));
103 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(int), o));
104 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &dg));
105 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &u));
106 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(int), n));
107 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &gd));
108 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(int), nb));
109 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &b));
110 CL_CHECK(clSetKernelArg(kernel, 9, sizeof(cl_mem), (void *) &bo));
111
112 CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) cmd_queue, kernel,
113 1, NULL, &global_item_size,
114 &local_item_size, 0, NULL, NULL));
115 CL_CHECK(clReleaseKernel(kernel));
116 }
117 break;
118 case GS_OP_MIN:
119 {
120 cl_kernel kernel = clCreateKernel(gs_program,
121 "gather_kernel_min", &err);
122 CL_CHECK(err);
123
124 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &v));
125 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(int), m));
126 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(int), o));
127 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &dg));
128 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &u));
129 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(int), n));
130 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &gd));
131 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(int), nb));
132 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &b));
133 CL_CHECK(clSetKernelArg(kernel, 9, sizeof(cl_mem), (void *) &bo));
134
135 CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) cmd_queue, kernel,
136 1, NULL, &global_item_size,
137 &local_item_size, 0, NULL, NULL));
138 CL_CHECK(clReleaseKernel(kernel));
139 }
140 break;
141 case GS_OP_MAX:
142 {
143 cl_kernel kernel = clCreateKernel(gs_program,
144 "gather_kernel_max", &err);
145 CL_CHECK(err);
146
147 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &v));
148 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(int), m));
149 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(int), o));
150 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &dg));
151 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &u));
152 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(int), n));
153 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &gd));
154 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(int), nb));
155 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &b));
156 CL_CHECK(clSetKernelArg(kernel, 9, sizeof(cl_mem), (void *) &bo));
157
158 CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) cmd_queue, kernel,
159 1, NULL, &global_item_size,
160 &local_item_size, 0, NULL, NULL));
161 CL_CHECK(clReleaseKernel(kernel));
162 }
163 break;
164 }
165}
166
170void opencl_scatter_kernel(void *v, int *m, void *dg,
171 void *u, int *n, void *gd,
172 int *nb, void *b, void *bo, void *cmd_queue) {
173 cl_int err;
174
175 if (gs_program == NULL)
176 opencl_kernel_jit(gs_kernels, (cl_program *) &gs_program);
177
178 cl_kernel kernel = clCreateKernel(gs_program, "scatter_kernel", &err);
179 CL_CHECK(err);
180
181 CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &v));
182 CL_CHECK(clSetKernelArg(kernel, 1, sizeof(int), m));
183 CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &dg));
184 CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &u));
185 CL_CHECK(clSetKernelArg(kernel, 4, sizeof(int), n));
186 CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &gd));
187 CL_CHECK(clSetKernelArg(kernel, 6, sizeof(int), nb));
188 CL_CHECK(clSetKernelArg(kernel, 7, sizeof(cl_mem), (void *) &b));
189 CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &bo));
190
191 const int nblks = ((*m) + 256 - 1) / 256;
192 const size_t global_item_size = 256 * nblks;
193 const size_t local_item_size = 256;
194
195 CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) cmd_queue, kernel, 1,
196 NULL, &global_item_size, &local_item_size,
197 0, NULL, NULL));
198 CL_CHECK(clReleaseKernel(kernel));
199}
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
u
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ u
Definition ax_helm_full_kernel.h:46
v
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ const T *__restrict__ v
Definition ax_helm_full_kernel.h:47
device_config.h
GS_OP_MAX
#define GS_OP_MAX
Definition gs.c:53
GS_OP_ADD
#define GS_OP_ADD
Definition gs.c:50
opencl_gather_kernel
void opencl_gather_kernel(void *v, int *m, int *o, void *dg, void *u, int *n, void *gd, int *nb, void *b, void *bo, int *op, void *cmd_queue)
Definition gs.c:58
GS_OP_MUL
#define GS_OP_MUL
Definition gs.c:51
GS_OP_MIN
#define GS_OP_MIN
Definition gs.c:52
opencl_scatter_kernel
void opencl_scatter_kernel(void *v, int *m, void *dg, void *u, int *n, void *gd, int *nb, void *b, void *bo, void *cmd_queue)
Definition gs.c:170
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
gs_program
void * gs_program