Neko 1.99.1
A portable framework for high-order spectral element flow simulations
Loading...
Searching...
No Matches
opr_convect_scalar.cu
Go to the documentation of this file.
1/*
2 Copyright (c) 2021-2023, 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#include <string.h>
36#include <stdlib.h>
37#include <stdio.h>
38#include "convect_scalar_kernel.h"
39#include <device/device_config.h>
40#include <device/cuda/check.h>
41
42extern "C" {
43 #include <common/neko_log.h>
44}
45
46template < const int >
47int tune_convect_scalar(void *du, void *u,
48 void *cr, void *cs, void *ct,
49 void *dx, void *dy, void *dz,
50 int *nel, int *lx);
51
52extern "C" {
53
57 void cuda_convect_scalar(void *du, void *u,
58 void *cr, void *cs, void *ct,
59 void *dx, void *dy, void *dz,
60 int *nel, int *lx) {
61
62 static int autotune[17] = { 0 };
63
64 const dim3 nthrds_1d(1024, 1, 1);
65 const dim3 nthrds_kstep((*lx), (*lx), 1);
66 const dim3 nblcks((*nel), 1, 1);
67 const cudaStream_t stream = (cudaStream_t) glb_cmd_queue;
68
69#define CASE_1D(LX) \
70 convect_scalar_kernel_1d<real, LX, 1024> \
71 <<<nblcks, nthrds_1d, 0, stream>>> \
72 ((real *) du, (real *) u, \
73 (real *) cr, (real *) cs, (real *) ct, \
74 (real *) dx, (real *) dy, (real *) dz); \
75 CUDA_CHECK(cudaGetLastError());
76
77#define CASE_KSTEP(LX) \
78 convect_scalar_kernel_kstep<real, LX> \
79 <<<nblcks, nthrds_kstep, 0, stream>>> \
80 ((real *) du, (real *) u, \
81 (real *) cr, (real *) cs, (real *) ct, \
82 (real *) dx, (real *) dy, (real *) dz); \
83 CUDA_CHECK(cudaGetLastError());
84
85#define CASE(LX) \
86 case LX: \
87 if(autotune[LX] == 0 ) { \
88 autotune[LX]=tune_convect_scalar<LX>(du, u, \
89 cr, cs, ct, \
90 dx, dy, dz, \
91 nel, lx); \
92 } else if (autotune[LX] == 1 ) { \
93 CASE_1D(LX); \
94 } else if (autotune[LX] == 2 ) { \
95 CASE_KSTEP(LX); \
96 } \
97 break
98
99#define CASE_LARGE(LX) \
100 case LX: \
101 CASE_KSTEP(LX); \
102 break
103
104
105 if ((*lx) < 11) {
106 switch(*lx) {
107 CASE(2);
108 CASE(3);
109 CASE(4);
110 CASE(5);
111 CASE(6);
112 CASE(7);
113 CASE(8);
114 CASE(9);
115 CASE(10);
116 default:
117 {
118 fprintf(stderr, __FILE__ ": size not supported: %d\n", *lx);
119 exit(1);
120 }
121 }
122 }
123 else {
124 switch(*lx) {
125 CASE_LARGE(11);
126 CASE_LARGE(12);
127 CASE_LARGE(13);
128 CASE_LARGE(14);
129 CASE_LARGE(15);
130 CASE_LARGE(16);
131 default:
132 {
133 fprintf(stderr, __FILE__ ": size not supported: %d\n", *lx);
134 exit(1);
135 }
136 }
137 }
138 }
139}
140
141template < const int LX >
142int tune_convect_scalar(void *du, void *u,
143 void *cr, void *cs, void *ct,
144 void *dx, void *dy, void *dz,
145 int *nel, int *lx) {
146 cudaEvent_t start,stop;
147 float time1,time2;
148 int retval;
149
150 const dim3 nthrds_1d(1024, 1, 1);
151 const dim3 nthrds_kstep((*lx), (*lx), 1);
152 const dim3 nblcks((*nel), 1, 1);
153 const cudaStream_t stream = (cudaStream_t) glb_cmd_queue;
154
155 char *env_value = NULL;
156 char neko_log_buf[80];
157
158 env_value=getenv("NEKO_AUTOTUNE");
159
160 sprintf(neko_log_buf, "Autotune convect_scalar (lx: %d)", *lx);
161 log_section(neko_log_buf);
162
163 if(env_value) {
164 if( !strcmp(env_value,"1D") ) {
165 CASE_1D(LX);
166 sprintf(neko_log_buf,"Set by env : 1 (1D)");
167 log_message(neko_log_buf);
168 log_end_section();
169 return 1;
170 } else if( !strcmp(env_value,"KSTEP") ) {
171 CASE_KSTEP(LX);
172 sprintf(neko_log_buf,"Set by env : 2 (KSTEP)");
173 log_message(neko_log_buf);
174 log_end_section();
175 return 2;
176 } else {
177 sprintf(neko_log_buf, "Invalid value set for NEKO_AUTOTUNE");
178 log_error(neko_log_buf);
179 }
180 }
181
182 cudaEventCreate(&start);
183 cudaEventCreate(&stop);
184
185 cudaEventRecord(start,0);
186
187 for(int i = 0; i < 100; i++) {
188 CASE_1D(LX);
189 }
190
191 cudaEventRecord(stop,0);
192 cudaEventSynchronize(stop);
193 cudaEventElapsedTime(&time1, start, stop);
194
195 cudaEventRecord(start,0);
196
197 for(int i = 0; i < 100; i++) {
198 CASE_KSTEP(LX);
199 }
200
201 cudaEventRecord(stop,0);
202 cudaEventSynchronize(stop);
203 cudaEventElapsedTime(&time2, start, stop);
204
205 if(time1 < time2) {
206 retval = 1;
207 } else {
208 retval = 2;
209 }
210
211 sprintf(neko_log_buf, "Chose : %d (%s)", retval,
212 (retval > 1 ? "KSTEP" : "1D"));
213 log_message(neko_log_buf);
214 log_end_section();
215 return retval;
216}
i
const int i
Definition ax_helm_full_kernel.h:95
u
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ u
Definition ax_helm_full_kernel.h:46
dx
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dx
Definition ax_helm_full_kernel.h:49
dz
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dz
Definition ax_helm_full_kernel.h:51
dy
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dy
Definition ax_helm_full_kernel.h:50
cr
__global__ void const T *__restrict__ const T *__restrict__ cr
Definition convect_scalar_kernel.h:112
cs
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ cs
Definition convect_scalar_kernel.h:113
ct
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ ct
Definition convect_scalar_kernel.h:114
dirichlet_apply_scalar_kernel
__global__ void dirichlet_apply_scalar_kernel(const int *__restrict__ msk, T *__restrict__ x, const T g, const int m)
Definition dirichlet_kernel.h:42
device_config.h
convect_scalar_kernel.h
neko_log.h
log_error
void log_error(char *msg)
log_message
void log_message(char *msg)
log_end_section
void log_end_section()
log_section
void log_section(char *msg)
CASE
#define CASE(LX)
CASE_KSTEP
#define CASE_KSTEP(LX)
CASE_LARGE
#define CASE_LARGE(LX)
CASE_1D
#define CASE_1D(LX)
tune_convect_scalar
int tune_convect_scalar(void *du, void *u, void *cr, void *cs, void *ct, void *dx, void *dy, void *dz, int *nel, int *lx)
Definition opr_convect_scalar.cu:142
cuda_convect_scalar
void cuda_convect_scalar(void *du, void *u, void *cr, void *cs, void *ct, void *dx, void *dy, void *dz, int *nel, int *lx)
Definition opr_convect_scalar.cu:57