Squash SYCL DeviceContext and DeviceStream
[alexxy/gromacs.git] / src / gromacs / ewald / CMakeLists.txt
1 #
2 # This file is part of the GROMACS molecular simulation package.
3 #
4 # Copyright (c) 2014,2015,2016,2017,2018 by the GROMACS development team.
5 # Copyright (c) 2019,2020, by the GROMACS development team, led by
6 # Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
7 # and including many others, as listed in the AUTHORS file in the
8 # top-level source directory and at http://www.gromacs.org.
9 #
10 # GROMACS is free software; you can redistribute it and/or
11 # modify it under the terms of the GNU Lesser General Public License
12 # as published by the Free Software Foundation; either version 2.1
13 # of the License, or (at your option) any later version.
14 #
15 # GROMACS is distributed in the hope that it will be useful,
16 # but WITHOUT ANY WARRANTY; without even the implied warranty of
17 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18 # Lesser General Public License for more details.
19 #
20 # You should have received a copy of the GNU Lesser General Public
21 # License along with GROMACS; if not, see
22 # http://www.gnu.org/licenses, or write to the Free Software Foundation,
23 # Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
24 #
25 # If you want to redistribute modifications to GROMACS, please
26 # consider that scientific software is very special. Version
27 # control is crucial - bugs must be traceable. We will be happy to
28 # consider code for inclusion in the official distribution, but
29 # derived work must not be called official GROMACS. Details are found
30 # in the README & COPYING files - if they are missing, get the
31 # official version at http://www.gromacs.org.
32 #
33 # To help us fund GROMACS development, we humbly ask that you cite
34 # the research papers on the package. Check out http://www.gromacs.org.
35
36 gmx_add_libgromacs_sources(
37     calculate_spline_moduli.cpp
38     ewald.cpp
39     ewald_utils.cpp
40     long_range_correction.cpp
41     pme.cpp
42     pme_gather.cpp
43     pme_grid.cpp
44     pme_load_balancing.cpp
45     pme_only.cpp
46     pme_pp.cpp
47     pme_redistribute.cpp
48     pme_solve.cpp
49     pme_spline_work.cpp
50     pme_spread.cpp
51     # Files that implement stubs
52     pme_gpu_program.cpp
53     pme_pp_comm_gpu_impl.cpp
54     pme_coordinate_receiver_gpu_impl.cpp
55     pme_force_sender_gpu_impl.cpp
56     )
57 if (GMX_GPU_CUDA)
58     gmx_add_libgromacs_sources(
59         # CUDA-specific sources
60         pme_gather.cu
61         pme_gpu_3dfft.cu
62         pme_solve.cu
63         pme_spread.cu
64         pme_gpu_program_impl.cu
65         pme_pp_comm_gpu_impl.cu
66         pme_force_sender_gpu_impl.cu
67         pme_coordinate_receiver_gpu_impl.cu
68         # GPU-specific sources
69         pme_gpu.cpp
70         pme_gpu_internal.cpp
71         pme_gpu_timings.cpp
72         )
73 elseif (GMX_GPU_OPENCL)
74     gmx_add_libgromacs_sources(
75         # OpenCL-specific sources
76         pme_gpu_3dfft_ocl.cpp
77         pme_gpu_program_impl_ocl.cpp
78         # GPU-specific sources
79         pme_gpu.cpp
80         pme_gpu_internal.cpp
81         pme_gpu_timings.cpp
82         )
83 elseif (GMX_GPU_SYCL)
84     # SYCL-TODO: proper implementation
85     gmx_add_libgromacs_sources(
86         pme_gpu_program_impl.cpp
87         )
88     _gmx_add_files_to_property(SYCL_SOURCES
89         pme_gpu_program_impl.cpp
90         pme_gpu_program.cpp
91         )
92 else()
93     gmx_add_libgromacs_sources(
94         # Files that implement stubs
95         pme_gpu_program_impl.cpp
96         )
97 endif()
98
99 if (BUILD_TESTING)
100     add_subdirectory(tests)
101 endif()
102
103
104 set(PME_OCL_KERNEL_SOURCES
105     "${CMAKE_CURRENT_SOURCE_DIR}/pme_gpu_calculate_splines.clh"
106     "${CMAKE_CURRENT_SOURCE_DIR}/pme_solve.clh"
107     "${CMAKE_CURRENT_SOURCE_DIR}/pme_gather.clh"
108     "${CMAKE_CURRENT_SOURCE_DIR}/pme_spread.clh")
109
110 if(CLANG_TIDY_EXE)
111    set(OCL_COMPILER "${CLANG_TIDY_EXE}")
112    set(CLANG_TIDY_ARGS "-quiet;-checks=*,-readability-implicit-bool-conversion,-llvm-header-guard,-hicpp-signed-bitwise,-clang-analyzer-deadcode.DeadStores,-google-readability-todo,-clang-diagnostic-padded,-fcomment-block-commands=internal;--;${CMAKE_C_COMPILER}")
113 else()
114    set(OCL_COMPILER "${CMAKE_C_COMPILER}")
115 endif()
116
117 # TODO: test all warp sizes on all vendor targets?
118 foreach(VENDOR AMD NVIDIA INTEL)
119     foreach(WARPSIZE 16 32 64)
120         math(EXPR SPREAD_WG_SIZE "8*${WARPSIZE}")
121         math(EXPR SOLVE_WG_SIZE "8*${WARPSIZE}")
122         math(EXPR GATHER_WG_SIZE "4*${WARPSIZE}")
123         set(OBJ_FILE pme_ocl_kernel_warpSize${WARPSIZE}_${VENDOR}.o)
124         add_custom_command(OUTPUT ${OBJ_FILE} COMMAND ${OCL_COMPILER}
125         ${CMAKE_CURRENT_SOURCE_DIR}/pme_program.cl ${CLANG_TIDY_ARGS}
126         -Xclang -finclude-default-header  -D_${VENDOR}_SOURCE_
127         -Dwarp_size=${WARPSIZE}
128         -Dorder=4
129         -DthreadsPerAtom=16
130         -Dc_pmeMaxUnitcellShift=2
131         -Dc_skipNeutralAtoms=false
132         -Dc_virialAndEnergyCount=7
133         -Dc_spreadWorkGroupSize=${SPREAD_WG_SIZE}
134         -Dc_solveMaxWorkGroupSize=${SOLVE_WG_SIZE}
135         -Dc_gatherWorkGroupSize=${GATHER_WG_SIZE}
136         -DDIM=3 -DXX=0 -DYY=1 -DZZ=2
137         -DwrapX=true -DwrapY=true
138         -c -I ${CMAKE_SOURCE_DIR}/src -std=cl1.2
139         -Weverything  -Wno-conversion -Wno-missing-variable-declarations -Wno-used-but-marked-unused
140         -Wno-cast-align -Wno-incompatible-pointer-types
141         # to avoid  "warning: unknown command tag name" for \internal
142         -Wno-documentation-unknown-command
143         # to avoid pme_gpu_types.h:100:52: warning: padding struct 'struct PmeGpuConstParams' with 4 bytes to align 'd_virialAndEnergy'
144         -Wno-padded
145         -o${OBJ_FILE}
146         )
147         list(APPEND PME_OCL_KERNELS ${OBJ_FILE})
148     endforeach()
149 endforeach()
150
151 add_custom_target(ocl_pme_kernels DEPENDS ${PME_OCL_KERNELS} )
152 gmx_set_custom_target_output(ocl_pme_kernels ${PME_OCL_KERNELS})