Neko  0.9.99
A portable framework for high-order spectral element flow simulations
coef.hip
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 #include <stdio.h>
36 #include <hip/hip_runtime.h>
37 #include <device/device_config.h>
38 #include <device/hip/check.h>
39 #include "coef_kernel.h"
40 
41 extern "C" {
42 
46  void hip_coef_generate_geo(void *G11, void *G12, void *G13,
47  void *G22, void *G23, void *G33,
48  void *drdx, void *drdy, void *drdz,
49  void *dsdx, void *dsdy, void *dsdz,
50  void *dtdx, void *dtdy, void *dtdz,
51  void *jacinv, void *w3, int *nel,
52  int *lx, int *gdim) {
53 
54  const dim3 nthrds(1024, 1, 1);
55  const dim3 nblcks((*nel), 1, 1);
56 
57 #define GEO_CASE(LX) \
58  case LX: \
59  hipLaunchKernelGGL( \
60  HIP_KERNEL_NAME(coef_generate_geo_kernel<real, LX, 1024>), \
61  nblcks, nthrds, 0, (hipStream_t) glb_cmd_queue, \
62  (real *) G11, (real *) G12, (real *) G13, \
63  (real *) G22, (real *) G23, (real *) G33, \
64  (real *) drdx, (real *) drdy, (real *) drdz, \
65  (real *) dsdx, (real *) dsdy, (real *) dsdz, \
66  (real *) dtdx, (real *) dtdy, (real *) dtdz, \
67  (real *) jacinv, (real *) w3, *gdim); \
68  HIP_CHECK(hipGetLastError()); \
69  break
70 
71  switch(*lx) {
72  GEO_CASE(2);
73  GEO_CASE(3);
74  GEO_CASE(4);
75  GEO_CASE(5);
76  GEO_CASE(6);
77  GEO_CASE(7);
78  GEO_CASE(8);
79  GEO_CASE(9);
80  GEO_CASE(10);
81  GEO_CASE(11);
82  GEO_CASE(12);
83  GEO_CASE(13);
84  GEO_CASE(14);
85  GEO_CASE(15);
86  GEO_CASE(16);
87  default:
88  {
89  fprintf(stderr, __FILE__ ": size not supported: %d\n", *lx);
90  exit(1);
91  }
92  }
93  }
94 
98  void hip_coef_generate_dxyzdrst(void *drdx, void *drdy, void *drdz,
99  void *dsdx, void *dsdy, void *dsdz,
100  void *dtdx, void *dtdy, void *dtdz,
101  void *dxdr, void *dydr, void *dzdr,
102  void *dxds, void *dyds, void *dzds,
103  void *dxdt, void *dydt, void *dzdt,
104  void *dx, void *dy, void *dz,
105  void *x, void *y, void *z,
106  void *jacinv, void *jac,
107  int *lx, int *nel) {
108 
109  const int n = (*nel) * (*lx) * (*lx) * (*lx);
110  const dim3 nthrds(1024, 1, 1);
111  const dim3 nblcks_dxyz((*nel), 1, 1);
112  const dim3 nblcks_drst((n + 1024 - 1)/ 1024, 1, 1);
113 
114 #define DXYZDRST_CASE(LX) \
115  case LX: \
116  hipLaunchKernelGGL( \
117  HIP_KERNEL_NAME(coef_generate_dxyz_kernel<real, LX, 1024>), \
118  nblcks_dxyz, nthrds, 0, (hipStream_t) glb_cmd_queue, \
119  (real *) dxdr, (real *) dydr, (real *) dzdr, \
120  (real *) dxds, (real *) dyds, (real *) dzds, \
121  (real *) dxdt, (real *) dydt, (real *) dzdt, \
122  (real *) dx, (real *) dy, (real *) dz, \
123  (real *) x, (real *) y, (real *) z); \
124  HIP_CHECK(hipGetLastError()); \
125  break
126 
127  switch(*lx) {
128  DXYZDRST_CASE(2);
129  DXYZDRST_CASE(3);
130  DXYZDRST_CASE(4);
131  DXYZDRST_CASE(5);
132  DXYZDRST_CASE(6);
133  DXYZDRST_CASE(7);
134  DXYZDRST_CASE(8);
135  DXYZDRST_CASE(9);
136  DXYZDRST_CASE(10);
137  DXYZDRST_CASE(11);
138  DXYZDRST_CASE(12);
139  DXYZDRST_CASE(13);
140  DXYZDRST_CASE(14);
141  DXYZDRST_CASE(15);
142  DXYZDRST_CASE(16);
143  default:
144  {
145  fprintf(stderr, __FILE__ ": size not supported: %d\n", *lx);
146  exit(1);
147  }
148  }
149 
150  hipLaunchKernelGGL(HIP_KERNEL_NAME(coef_generate_drst_kernel<real>),
151  nblcks_drst, nthrds, 0, (hipStream_t) glb_cmd_queue,
152  (real *) jac, (real *) jacinv,
153  (real *) drdx, (real *) drdy, (real *) drdz,
154  (real *) dsdx, (real *) dsdy, (real *) dsdz,
155  (real *) dtdx, (real *) dtdy, (real *) dtdz,
156  (real *) dxdr, (real *) dydr, (real *) dzdr,
157  (real *) dxds, (real *) dyds, (real *) dzds,
158  (real *) dxdt, (real *) dydt, (real *) dzdt, n);
159  HIP_CHECK(hipGetLastError());
160 
161  }
162 }
hip_coef_generate_geo
void hip_coef_generate_geo(void *G11, void *G12, void *G13, void *G22, void *G23, void *G33, void *drdx, void *drdy, void *drdz, void *dsdx, void *dsdy, void *dsdz, void *dtdx, void *dtdy, void *dtdz, void *jacinv, void *w3, int *nel, int *lx, int *gdim)
Definition: coef.hip:46
hip_coef_generate_dxyzdrst
void hip_coef_generate_dxyzdrst(void *drdx, void *drdy, void *drdz, void *dsdx, void *dsdy, void *dsdz, void *dtdx, void *dtdy, void *dtdz, void *dxdr, void *dydr, void *dzdr, void *dxds, void *dyds, void *dzds, void *dxdt, void *dydt, void *dzdt, void *dx, void *dy, void *dz, void *x, void *y, void *z, void *jacinv, void *jac, int *lx, int *nel)
Definition: coef.hip:98
GEO_CASE
#define GEO_CASE(LX)
DXYZDRST_CASE
#define DXYZDRST_CASE(LX)
drdy
__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__ const T *__restrict__ const T *__restrict__ drdy
Definition: ax_helm_full_kernel.h:54
drdz
__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__ const T *__restrict__ const T *__restrict__ const T *__restrict__ drdz
Definition: ax_helm_full_kernel.h:55
dsdz
__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__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dsdz
Definition: ax_helm_full_kernel.h:58
dsdy
__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__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dsdy
Definition: ax_helm_full_kernel.h:57
dtdy
__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__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dtdy
Definition: ax_helm_full_kernel.h:60
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
drdx
__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__ const T *__restrict__ drdx
Definition: ax_helm_full_kernel.h:53
dtdz
__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__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dtdz
Definition: ax_helm_full_kernel.h:61
dsdx
__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__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dsdx
Definition: ax_helm_full_kernel.h:56
dtdx
__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__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dtdx
Definition: ax_helm_full_kernel.h:59
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
jacinv
__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__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ jacinv
Definition: ax_helm_full_kernel.h:62
x
__global__ void const T *__restrict__ x
Definition: cdtp_kernel.h:106
w3
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ w3
Definition: cdtp_kernel.h:113
device_config.h
real
double real
Definition: device_config.h:12
glb_cmd_queue
void * glb_cmd_queue
HIP_CHECK
#define HIP_CHECK(err)
Definition: check.h:8
coef_kernel.h