Neko  0.9.0
A portable framework for high-order spectral element flow simulations
device_coef.F90
Go to the documentation of this file.
1 ! Copyright (c) 2022-2024, The Neko Authors
2 ! All rights reserved.
3 !
4 ! Redistribution and use in source and binary forms, with or without
5 ! modification, are permitted provided that the following conditions
6 ! are met:
7 !
8 ! * Redistributions of source code must retain the above copyright
9 ! notice, this list of conditions and the following disclaimer.
10 !
11 ! * Redistributions in binary form must reproduce the above
12 ! copyright notice, this list of conditions and the following
13 ! disclaimer in the documentation and/or other materials provided
14 ! with the distribution.
15 !
16 ! * Neither the name of the authors nor the names of its
17 ! contributors may be used to endorse or promote products derived
18 ! from this software without specific prior written permission.
19 !
20 ! THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 ! "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 ! LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 ! FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 ! COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 ! INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 ! BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 ! LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
28 ! CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 ! LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30 ! ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31 ! POSSIBILITY OF SUCH DAMAGE.
32 !
34  use num_types, only : rp
35  use utils, only : neko_error
36  use, intrinsic :: iso_c_binding, only : c_ptr, c_int
37  implicit none
38  private
39 
41 
42 #ifdef HAVE_HIP
43  interface
44  subroutine hip_coef_generate_geo(G11, G12, G13, G22, G23, G33, &
45  drdx, drdy, drdz, dsdx, dsdy, dsdz, dtdx, dtdy, dtdz, &
46  jacinv, w3, nel, lx, gdim) &
47  bind(c, name='hip_coef_generate_geo')
48  use, intrinsic :: iso_c_binding
49  implicit none
50  type(c_ptr), value :: G11, G12, G13, G22, G23, G33
51  type(c_ptr), value :: drdx, drdy, drdz
52  type(c_ptr), value :: dsdx, dsdy, dsdz
53  type(c_ptr), value :: dtdx, dtdy, dtdz
54  type(c_ptr), value :: jacinv, w3
55  integer(c_int) :: nel, gdim, lx
56  end subroutine hip_coef_generate_geo
57  end interface
58 
59  interface
60  subroutine hip_coef_generate_dxyzdrst(drdx, drdy, drdz, dsdx, dsdy, &
61  dsdz, dtdx, dtdy, dtdz, dxdr, dydr, dzdr, dxds, dyds, dzds, dxdt, &
62  dydt, dzdt, dx, dy, dz, x, y, z, jacinv, jac, lx, nel) &
63  bind(c, name='hip_coef_generate_dxyzdrst')
64  use, intrinsic :: iso_c_binding
65  implicit none
66  type(c_ptr), value :: drdx, drdy, drdz
67  type(c_ptr), value :: dsdx, dsdy, dsdz
68  type(c_ptr), value :: dtdx, dtdy, dtdz
69  type(c_ptr), value :: dxdr, dydr, dzdr
70  type(c_ptr), value :: dxds, dyds, dzds
71  type(c_ptr), value :: dxdt, dydt, dzdt
72  type(c_ptr), value :: dx, dy, dz, x, y, z
73  type(c_ptr), value :: jacinv, jac
74  integer(c_int) :: lx, nel
75  end subroutine hip_coef_generate_dxyzdrst
76  end interface
77 #elif HAVE_CUDA
78  interface
79  subroutine cuda_coef_generate_geo(G11, G12, G13, G22, G23, G33, &
80  drdx, drdy, drdz, dsdx, dsdy, dsdz, dtdx, dtdy, dtdz, &
81  jacinv, w3, nel, lx, gdim) &
82  bind(c, name='cuda_coef_generate_geo')
83  use, intrinsic :: iso_c_binding
84  implicit none
85  type(c_ptr), value :: G11, G12, G13, G22, G23, G33
86  type(c_ptr), value :: drdx, drdy, drdz
87  type(c_ptr), value :: dsdx, dsdy, dsdz
88  type(c_ptr), value :: dtdx, dtdy, dtdz
89  type(c_ptr), value :: jacinv, w3
90  integer(c_int) :: nel, gdim, lx
91  end subroutine cuda_coef_generate_geo
92  end interface
93 
94  interface
95  subroutine cuda_coef_generate_dxyzdrst(drdx, drdy, drdz, dsdx, dsdy, &
96  dsdz, dtdx, dtdy, dtdz, dxdr, dydr, dzdr, dxds, dyds, dzds, dxdt, &
97  dydt, dzdt, dx, dy, dz, x, y, z, jacinv, jac, lx, nel) &
98  bind(c, name='cuda_coef_generate_dxyzdrst')
99  use, intrinsic :: iso_c_binding
100  implicit none
101  type(c_ptr), value :: drdx, drdy, drdz
102  type(c_ptr), value :: dsdx, dsdy, dsdz
103  type(c_ptr), value :: dtdx, dtdy, dtdz
104  type(c_ptr), value :: dxdr, dydr, dzdr
105  type(c_ptr), value :: dxds, dyds, dzds
106  type(c_ptr), value :: dxdt, dydt, dzdt
107  type(c_ptr), value :: dx, dy, dz, x, y, z
108  type(c_ptr), value :: jacinv, jac
109  integer(c_int) :: lx, nel
110  end subroutine cuda_coef_generate_dxyzdrst
111  end interface
112 #elif HAVE_OPENCL
113  interface
114  subroutine opencl_coef_generate_geo(G11, G12, G13, G22, G23, G33, &
115  drdx, drdy, drdz, dsdx, dsdy, dsdz, dtdx, dtdy, dtdz, &
116  jacinv, w3, nel, lx, gdim) &
117  bind(c, name='opencl_coef_generate_geo')
118  use, intrinsic :: iso_c_binding
119  implicit none
120  type(c_ptr), value :: G11, G12, G13, G22, G23, G33
121  type(c_ptr), value :: drdx, drdy, drdz
122  type(c_ptr), value :: dsdx, dsdy, dsdz
123  type(c_ptr), value :: dtdx, dtdy, dtdz
124  type(c_ptr), value :: jacinv, w3
125  integer(c_int) :: nel, gdim, lx
126  end subroutine opencl_coef_generate_geo
127  end interface
128 
129  interface
130  subroutine opencl_coef_generate_dxyzdrst(drdx, drdy, drdz, dsdx, dsdy, &
131  dsdz, dtdx, dtdy, dtdz, dxdr, dydr, dzdr, dxds, dyds, dzds, dxdt, &
132  dydt, dzdt, dx, dy, dz, x, y, z, jacinv, jac, lx, nel) &
133  bind(c, name='opencl_coef_generate_dxyzdrst')
134  use, intrinsic :: iso_c_binding
135  implicit none
136  type(c_ptr), value :: drdx, drdy, drdz
137  type(c_ptr), value :: dsdx, dsdy, dsdz
138  type(c_ptr), value :: dtdx, dtdy, dtdz
139  type(c_ptr), value :: dxdr, dydr, dzdr
140  type(c_ptr), value :: dxds, dyds, dzds
141  type(c_ptr), value :: dxdt, dydt, dzdt
142  type(c_ptr), value :: dx, dy, dz, x, y, z
143  type(c_ptr), value :: jacinv, jac
144  integer(c_int) :: lx, nel
145  end subroutine opencl_coef_generate_dxyzdrst
146  end interface
147 #endif
148 
149 contains
150 
151  subroutine device_coef_generate_geo(G11_d, G12_d, G13_d, G22_d, &
152  G23_d, G33_d, drdx_d, drdy_d, drdz_d, dsdx_d, dsdy_d, &
153  dsdz_d, dtdx_d, dtdy_d, dtdz_d, jacinv_d, w3_d, nel, lx, gdim)
154  type(c_ptr) :: g11_d, g12_d, g13_d, g22_d, g23_d, g33_d
155  type(c_ptr) :: drdx_d, drdy_d, drdz_d
156  type(c_ptr) :: dsdx_d, dsdy_d, dsdz_d
157  type(c_ptr) :: dtdx_d, dtdy_d, dtdz_d
158  type(c_ptr) :: jacinv_d, w3_d
159  integer :: nel, gdim, lx
160 
161 #ifdef HAVE_HIP
162  call hip_coef_generate_geo(g11_d, g12_d, g13_d, g22_d, g23_d, &
163  g33_d, drdx_d, drdy_d, drdz_d, dsdx_d, dsdy_d, dsdz_d, &
164  dtdx_d, dtdy_d, dtdz_d, jacinv_d, w3_d, nel, lx, gdim)
165 #elif HAVE_CUDA
166  call cuda_coef_generate_geo(g11_d, g12_d, g13_d, g22_d, g23_d, &
167  g33_d, drdx_d, drdy_d, drdz_d, dsdx_d, dsdy_d, dsdz_d, &
168  dtdx_d, dtdy_d, dtdz_d, jacinv_d, w3_d, nel, lx, gdim)
169 #elif HAVE_OPENCL
170  call opencl_coef_generate_geo(g11_d, g12_d, g13_d, g22_d, g23_d, &
171  g33_d, drdx_d, drdy_d, drdz_d, dsdx_d, dsdy_d, dsdz_d, &
172  dtdx_d, dtdy_d, dtdz_d, jacinv_d, w3_d, nel, lx, gdim)
173 #else
174  call neko_error('No device backend configured')
175 #endif
176 
177  end subroutine device_coef_generate_geo
178 
179  subroutine device_coef_generate_dxydrst(drdx_d, drdy_d, drdz_d, dsdx_d, dsdy_d,&
180  dsdz_d, dtdx_d, dtdy_d, dtdz_d, dxdr_d, dydr_d, dzdr_d, dxds_d, &
181  dyds_d, dzds_d, dxdt_d, dydt_d, dzdt_d, dx_d, dy_d, dz_d, &
182  x_d, y_d, z_d, jacinv_d, jac_d, lx, nel)
183  type(c_ptr) :: drdx_d, drdy_d, drdz_d
184  type(c_ptr) :: dsdx_d, dsdy_d, dsdz_d
185  type(c_ptr) :: dtdx_d, dtdy_d, dtdz_d
186  type(c_ptr) :: dxdr_d, dydr_d, dzdr_d
187  type(c_ptr) :: dxds_d, dyds_d, dzds_d
188  type(c_ptr) :: dxdt_d, dydt_d, dzdt_d
189  type(c_ptr) :: dx_d, dy_d, dz_d, x_d, y_d, z_d
190  type(c_ptr) :: jacinv_d, jac_d
191  integer :: lx, nel
192 
193 #ifdef HAVE_HIP
194  call hip_coef_generate_dxyzdrst(drdx_d, drdy_d, drdz_d, dsdx_d, &
195  dsdy_d, dsdz_d, dtdx_d, dtdy_d, dtdz_d, dxdr_d, dydr_d, &
196  dzdr_d, dxds_d, dyds_d, dzds_d, dxdt_d, dydt_d, dzdt_d, &
197  dx_d, dy_d, dz_d, x_d, y_d, z_d, jacinv_d, jac_d, lx, nel)
198 #elif HAVE_CUDA
199  call cuda_coef_generate_dxyzdrst(drdx_d, drdy_d, drdz_d, dsdx_d, &
200  dsdy_d, dsdz_d, dtdx_d, dtdy_d, dtdz_d, dxdr_d, dydr_d, &
201  dzdr_d, dxds_d, dyds_d, dzds_d, dxdt_d, dydt_d, dzdt_d, &
202  dx_d, dy_d, dz_d, x_d, y_d, z_d, jacinv_d, jac_d, lx, nel)
203 #elif HAVE_OPENCL
204  call opencl_coef_generate_dxyzdrst(drdx_d, drdy_d, drdz_d, dsdx_d, &
205  dsdy_d, dsdz_d, dtdx_d, dtdy_d, dtdz_d, dxdr_d, dydr_d, &
206  dzdr_d, dxds_d, dyds_d, dzds_d, dxdt_d, dydt_d, dzdt_d, &
207  dx_d, dy_d, dz_d, x_d, y_d, z_d, jacinv_d, jac_d, lx, nel)
208 #else
209  call neko_error('No device backend configured')
210 #endif
211  end subroutine device_coef_generate_dxydrst
212 
213 end module device_coef
void opencl_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.c:54
void opencl_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.c:125
void cuda_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.cu:97
void cuda_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.cu:45
subroutine, public device_coef_generate_dxydrst(drdx_d, drdy_d, drdz_d, dsdx_d, dsdy_d, dsdz_d, dtdx_d, dtdy_d, dtdz_d, dxdr_d, dydr_d, dzdr_d, dxds_d, dyds_d, dzds_d, dxdt_d, dydt_d, dzdt_d, dx_d, dy_d, dz_d, x_d, y_d, z_d, jacinv_d, jac_d, lx, nel)
subroutine, public device_coef_generate_geo(G11_d, G12_d, G13_d, G22_d, G23_d, G33_d, drdx_d, drdy_d, drdz_d, dsdx_d, dsdy_d, dsdz_d, dtdx_d, dtdy_d, dtdz_d, jacinv_d, w3_d, nel, lx, gdim)
integer, parameter, public rp
Global precision used in computations.
Definition: num_types.f90:12
Utilities.
Definition: utils.f90:35