- Using Spack branch e4s-22.05 based on Spack v0.18.0-dev
Sun May 15 06:59:11 2022 -0400
- Package added to E4S:
bricks
andbricks +cuda
- Replace
ninja
withpruners-ninja
- Release separate environments for NVIDIA, ROCm, and Intel GPU
-
Using Spack branch e4s-21.11 based on Spack v0.17.0
spack@v0.17.0
Fri Nov 5 17:11:18 2021 -0700
-
Packages added:
alquimia
butterflypack
catalyst
py-cinemasci
-
Using Spack 0.17.0's new ASP-based concretizer,
clingo
-
Updated
packages:
preferences in environment- Removed unnecessary package preferences
- Allow concretizer to select latest versions
- Prefer
cuda@11.4.2
for x86_64 model environment - Prefer
cuda@11.0.2
for ppc64le model environment
-
CUDA compute capability
- x86_64: build for
cuda_arch=80
(A100) - ppc64le: build for
cuda_arch=70
(V100)
- x86_64: build for
-
AMD: building for
gfx908
(MI100) -
Enabled additional
+cuda
variantsflecsi +cuda
heffte +cuda
petsc +cuda
slepc +cuda
trilinos +cuda
(introduced via newly-releasedtrilinos@13.2.0
upcxx +cuda
(introduced via newly-releasedupcxx@2021.9.0
)vtk-m +cuda
-
Enabled additional
+rocm
variantsamrex +rocm
ginkgo +rocm
heffte +rocm
magma +rocm
mfem +rocm
slate +rocm
strumpack +rocm
-
Uses
ROCM 4.3.1
stack
For more details see E4S 21.11 release notes
-
Using Spack branch e4s-21.08
spack@develop
- Branched from commit f444303ce5c1fd274192dbc10e6783ddd601e0bd
Tue Aug 17 16:19:31 2021 +0200
-
Packages added:
charliecloud
datatransferkit
flit
flux-core
geopm
nvhpc
PaRSEC
(+cuda and ~cuda)phist
veloc
vtk-m
-
Packages commented out due to build issues:
STRUMPACK +rocm
(Issue #25433)
-
HDF5 1.12.0
is being used by a number of packages that were previously using HDF 1.10.7 -
OpenBLAS 0.3.17
is now the preferred BLAS provider where previously it was version 0.3.10 -
Trilinos
is being used with most variants turned on, thereby requiring only a single install of trilinos where previously three were required to satisfy the environment's concretization