ca76ff360f
augment cmake library search path to include the CUDA stubs library folder
...
this will help configuring and compiling LAMMPS with CUDA support on
machines where there is no CUDA driver installed
2022-03-10 21:17:58 -05:00
af8d1bd768
fix typo
2022-01-30 09:00:39 -05:00
4b22962ec1
roll back MPI target alias change one more time. must wait until CMake 3.11 or later is minimum
2022-01-30 08:47:46 -05:00
dc0e013297
create alias target MPI::ANY_CXX to be used instead of MPI::MPI_CXX
...
With this alias it is possible to transparently refer to either the
real imported MPI library or to the MPI stub library. this further
reduced the need for if statements related to MPI. Some uses of
MPI::MPI_CXX remain but they are all in branches of the script code
where BUILD_MPI is enabled and thus the imported target will be present.
2022-01-30 07:48:16 -05:00
d6fa3a08cd
roll back library aliasing changes as that negatively interferes with unit tests
2022-01-28 21:54:32 -05:00
db4e69bf38
further simplification by making STUBS an interface and alias to MPI::MPI_CXX
2022-01-28 01:00:37 -05:00
62ff23abe7
simplify building serial executables by incorporating the STUBS/mpi.o object in LAMMPS library
2022-01-27 15:59:48 -05:00
053d915fc4
drop -ffast-math for HIP also when compiling with CMake
2021-12-27 20:14:30 -05:00
f5fa892ec8
Merge pull request #2916 from rbberger/rocm_updates
...
Updates to support ROCm 4.3 in GPU package
2021-09-22 16:23:19 -04:00
407f032a55
Update CMake variable descriptions
2021-09-22 15:14:39 -04:00
b1ebaa298c
build "fat" cuda binaries only with known toolkits
2021-09-14 17:17:38 -04:00
8db2d64f11
Updates to support ROCm 4.3 in GPU package
2021-08-31 17:56:01 -04:00
bc063f2e88
make formatting of if statements and string comparisons consistent
2021-05-25 15:06:08 -04:00
69d081cd6a
prefer system provided OpenCL library on macOS
2021-05-07 09:28:13 -04:00
4ccb4c9603
CMake: Add missing defines to GPU kernel compilation
2021-04-22 18:46:07 -04:00
717faa6515
correctly detect GPU package with CUDA api
2021-04-06 19:12:28 -04:00
00e4bdb2b7
Merge pull request #2692 from arghdos/master
...
Porting to new default platform for AMD/HIP in ROCm 4.1
2021-04-06 14:39:30 -04:00
88b9e99707
Update cmake/Modules/Packages/GPU.cmake
2021-04-05 15:26:58 -04:00
4c2fb7a431
Porting to new default platform for AMD/HIP in ROCm 4.1
2021-04-01 09:43:13 -05:00
d1b4af60a3
disable GPU binning via CUDPP by default and describe why
2021-03-16 11:58:47 -04:00
7d1670d91c
cuda 11.0 does not support sm_86, it was added in 11.1
2021-02-23 14:23:54 -05:00
4786391fad
must explicitly register fix gpu
2021-02-21 20:40:11 -05:00
0c6671ad64
do not always add styles that depend on other packages
2021-02-21 11:06:58 -05:00
d025b281cf
Build and link a static OpenCL loader library for all platforms
2021-02-21 01:28:44 -05:00
775446b60f
Add GERYON_NUMA_FISSION define in CMake
2021-02-16 15:03:17 -05:00
d85a5e3290
Remove OCL_TUNE option in CMake
...
The GPU package now auto-detects these settings.
2021-02-16 14:57:10 -05:00
224da33b22
Add missing fix_nh_gpu files to CMake build
2021-02-16 12:29:50 -05:00
06e484d3e0
make flags for supporting GPU archs consistent, add ampere consumer GPU arch
2021-02-07 15:30:55 -05:00
fc8b8d8825
disallow use of CUDPP with CUDA multiprocessor server
2021-02-05 16:44:26 -05:00
4758e43a2e
add support for setting -DLMP_GPU when the GPU package is installed
2021-01-11 10:18:09 -05:00
acb5fd2ebf
Simplify GenerateBinaryHeader utility
...
The function implementation was incomplete for handling the case
of multiple files. It would generate wrong C code in case more than
one files were specified. Since we don't have this use case, it's
safe to only leave the single input file implementation.
Closes issue #2294
2020-09-20 18:24:43 -04:00
5623009962
arch 3.0 is dropped in CUDA 11
2020-08-30 13:12:42 +03:00
aa82608142
print warning for unknown/unsupported CUDA versions
2020-07-15 11:28:33 -04:00
a082895b90
update CMake script for GPU package to include handling of Arch 8.0 and CUDA 11
2020-07-15 11:08:06 -04:00
723b9e6760
Try to keep compatibility with deprecated ROCm hcc compiler
2020-06-20 15:17:54 +03:00
2ea5d9956a
find_package is unnecessary for rocprim
2020-06-19 00:01:03 +03:00
d3baa414d2
Fix cmake for GPU ROCm 3.5 backend
2020-06-17 23:33:16 +03:00
6891fa2610
update a few URLs to use https: instead of http:
2020-06-10 12:34:09 -04:00
78f030a07a
test for *cubin.h files in lib/gpu that conflict with cmake compilation
2020-06-09 07:32:49 -04:00
1adda03202
cmake: only install shared lib and friends
2020-04-29 18:06:33 -06:00
ef3e071067
GPU.cmake: localize hackery
2020-04-28 07:43:31 -06:00
6be36196da
explicitly include STUBS MPI headers for serial compile since we seem to pick up the OpenMP ones otherwise
2020-04-28 01:36:21 -04:00
705c10ef47
restore original command
2020-04-28 01:12:49 -04:00
d3017744a9
updates for building with hip
2020-04-28 00:39:11 -04:00
29d667721a
tweak GPU.cmake so that hip_get_devices compiles and links
2020-04-27 22:29:41 -04:00
5542ad31dc
Merge branch 'master' into simplify_build_choices
...
# Conflicts:
# cmake/Modules/Packages/GPU.cmake
2020-04-11 21:31:07 -04:00
d2c4ade697
Add HIP_USE_DEVICE_SORT CMake option
2020-04-10 18:22:43 -04:00
4f8a1583b9
Restore --arch line removed during debugging
2020-04-10 14:25:46 -04:00
b7a2b375c6
Add includes and linked libs for HIP_PLATFORM=nvcc
...
Do not set CMAKE_CXX_COMPILER=hipcc in the NVCC case since it causes
all sorts of problems. Instead we use the regular host compiler and
set the necessary include and link paths.
2020-04-10 12:45:39 -04:00
6381940b9a
Attempt to get ROCm CUDA working with CMake
2020-04-09 13:53:41 -04:00
