Paul T. Bauman
d55a409bdb
Silence uninitialized var HIP warnings
...
This should be a benign change. What happens is that the first one or two
workitems/threads in each workgroup/block read a value and then broadcast
it (with __shfl_sync or similar) and then code branching happens
based on this value. But the compiler can't see it all the way
through it so we get some uninitialized var warnings.
2022-03-09 13:31:32 -06:00
Ruipeng Li
9c33e9a263
regression tests
2022-03-09 08:54:42 -08:00
Ruipeng Li
63c9fa65a2
add using hypre's spmv option
2022-03-08 22:11:31 -08:00
Wayne Mitchell
5e90f43cbc
astyle
2022-03-08 23:51:41 +00:00
Wayne Mitchell
34c16787b7
Addition of MatMat and TMatMat routines and clean up
2022-03-08 23:50:48 +00:00
Wayne Mitchell
297ff5d5a7
Par matmat is verified correcct for a small example
2022-03-08 21:08:47 +00:00
Ruipeng Li
7443a2ac6c
missed some sync in the last commit
2022-03-07 23:54:03 -08:00
Ruipeng Li
e1b9a56405
add gpu sync for mpi
2022-03-07 23:34:33 -08:00
Ruipeng Li
8ee20f4812
cudamallocasync
2022-03-07 16:54:56 -08:00
Ruipeng Li
df0f6dbba7
configure options: cublas; cudamallocasync
2022-03-07 16:40:32 -08:00
Ruipeng Li
d7728d0bce
updated ij driver for 2nd solve
2022-03-07 15:16:00 -08:00
Ruipeng Li
b97fbc13ed
sync device at ending timing
2022-03-07 15:13:08 -08:00
Ruipeng Li
c2e4836c1e
bug fix
2022-03-05 10:36:53 -08:00
Ruipeng Li
a51bb880a8
bug fix
2022-03-05 09:46:16 -08:00
Ruipeng Li
03546b428f
Merge branch 'master' of github.com:hypre-space/hypre into nvcollab
2022-03-04 22:18:02 -08:00
Paul T. Bauman
251cd3d269
Need -O1 instead of -O0 for HIP in debug mode ( #588 )
...
This PR changes -O0 in debug mode to -O1 with HIP (at this time).
2022-03-04 12:40:35 -08:00
Ruipeng Li
95e6433fc7
GPU support with single precision ( #572 )
...
This PR fixes the GPU support with single precision.
2022-03-04 12:05:32 -08:00
Ruipeng Li
ebd6eb88c3
bug fix; nonsquare rap ( #581 )
...
This PR fixes a corner case of the RAP routine for RAP matrix that is globally square but not locally.
2022-03-03 21:26:17 -08:00
Wayne Mitchell
d388a2766e
Lots of reorganization. This now has all functionality for par matmat and compiles, but needs debugging.
2022-03-01 01:59:27 +00:00
Wayne Mitchell
8112dd736f
Further cleanup and reorg of device_utils.c/h and addition of more functionality needed for par matmat
2022-02-19 00:40:27 +00:00
Paul T. Bauman
04af9a4cd9
HYPRE_Int -> HYPRE_BigInt ( #585 )
2022-02-18 12:16:35 -08:00
Wayne Mitchell
cda5b10a69
Single processor device rap works
2022-02-18 18:58:28 +00:00
Wayne Mitchell
47ae1c8a22
Start major reorganization of device_utils.h
2022-02-18 17:44:22 +00:00
Golam Rabbani
94070dd3a9
Updated CMakeLists.txt for SYCL ( #577 )
...
With CMake, enable CUDA stream by default when using SYCL.
2022-02-17 18:21:51 -08:00
Victor A. Paludetto Magri
33a5051398
Add SStruct IO functions ( #583 )
...
This PR adds support for native print/read functions of SStructMatrix and SStructVector. Other important changes are:
* Add public functions for reading StructMatrix and StructVector.
* Add a new set of regression tests called "io" to the TEST_sstruct folder.
2022-02-17 18:06:23 -08:00
Wayne Mitchell
1e289479f1
astyle
2022-02-17 00:56:49 +00:00
Ruipeng Li
9888903445
memory tracker
2022-02-16 15:04:20 -08:00
Ruipeng Li
c336122cc1
move to debug region
2022-02-16 14:19:15 -08:00
Victor A. Paludetto Magri
49dbf7b60a
Fix cross-compilation problem ( #580 )
...
This PR fixes issue #556 .
AC_CHECK_FILE was being used to test the existence of the .git folder. However, according to Autoconf manual, it does not work when cross-compiling. This PR implements another strategy for looking for the .git folder which works also when doing cross-compilation.
2022-02-16 07:55:02 -08:00
Wayne Mitchell
4136c63269
Switch to HYPRE_GPU_LAUNCH
2022-02-15 20:01:08 +00:00
Wayne Mitchell
7883708a4f
astyle
2022-02-15 18:05:12 +00:00
Wayne Mitchell
e3e99f3b9d
Add back include
2022-02-15 18:04:22 +00:00
Wayne Mitchell
1e2f711d01
Merge branch 'sycl' into sycl_par_matmat
2022-02-15 18:00:07 +00:00
Wayne Mitchell
e312ecbe31
Merge branch 'master' into sycl_par_matmat
2022-02-15 17:41:05 +00:00
Rui-peng Li
df0b1eac20
turn off print
2022-02-12 01:14:53 -06:00
Rui-peng Li
c430559482
remove fast hash
2022-02-12 01:09:15 -06:00
Rui-peng Li
c347b3c9e1
remove some changes
2022-02-12 00:53:55 -06:00
Ruipeng Li
7672c101cd
try to optimize hash
2022-02-11 22:17:47 -08:00
Ruipeng Li
c85a060d85
naive row nnz bound with spmv
2022-02-11 17:18:32 -08:00
Ulrike Yang
e5a82e81e6
specified SYCL support
2022-02-10 17:12:32 -08:00
Rob Falgout
ccd135d8da
Updating CHANGELOG
2022-02-10 10:18:32 -08:00
Rob Falgout
666f457d2b
Bumping release number to 2.24.0
2022-02-10 07:05:43 -08:00
Rob Falgout
4ee737b53c
Initial CHANGELOG update for new release
2022-02-10 07:02:26 -08:00
Ruipeng Li
9d99e1fb77
put unroll as a macro
2022-02-10 00:14:18 -05:00
Ruipeng Li
efed76505c
rename printf0; change unroll to 1
2022-02-09 23:44:49 -05:00
Wayne Mitchell
f8d482efad
Add packing on device with oneDPL for par matvec
2022-02-10 01:51:56 +00:00
Ruipeng Li
6180f1261c
minor change
2022-02-09 20:36:53 -05:00
Ruipeng Li
bef2862710
silence compiler warnings
2022-02-09 13:27:18 -05:00
Ruipeng Li
ab72d05bd8
Deviceomp ( #519 )
...
This PR fixes the build with Kokkos + OMP offload, supports OMP offload without linking CUDA libraries, and supports OMP offload on Intel GPUs.
2022-02-09 06:40:57 -08:00
Rui-peng Li
e0c24fd1f6
turn off printings
2022-02-08 16:34:05 -06:00
