zhangyiss/spack
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Trilinos launch blocking + maintainers (#49468)

Browse Source 
* Trilinos launch blocking + maintainers

Cuda launch blocking is not needed and slowing modern apps down. 

More maintainers to spot issues like this.

---------

Co-authored-by: psakievich <psakievich@users.noreply.github.com>
This commit is contained in:
psakievich 2025-03-14 02:08:30 -06:00 committed by GitHub
parent d09c5a4bd4
commit 8235aa1804
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 17 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
var/spack/repos/builtin/packages/trilinos/package.py
View File

@ -33,7 +33,17 @@ class Trilinos(CMakePackage, CudaPackage, ROCmPackage):
url = "https://github.com/trilinos/Trilinos/archive/refs/tags/trilinos-release-12-12-1.tar.gz"
git = "https://github.com/trilinos/Trilinos.git"
maintainers("keitat", "kuberry", "jwillenbring", "psakievich")
maintainers(
"keitat",
"kuberry",
"jwillenbring",
"psakievich",
"ccober6",
"fryeguy52",
"sebrowne",
"rppawlo",
"cgcgcg",
)
tags = ["e4s"]
@ -583,8 +593,8 @@ def url_for_version(self, version):
return url.format(version.dashed)
def setup_dependent_run_environment(self, env, dependent_spec):
if "+cuda" in self.spec:
# currently Trilinos doesn't perform the memory fence so
if self.spec.satisfies("@:13.1.0 +cuda"):
# older releases of Trilinos doesn't perform the memory fence so
# it relies on blocking CUDA kernel launch. This is needed
# in case the dependent app also run a CUDA backend via Trilinos
env.set("CUDA_LAUNCH_BLOCKING", "1")
@ -1086,7 +1096,8 @@ def setup_run_environment(self, env):
if "+exodus" in self.spec:
env.prepend_path("PYTHONPATH", self.prefix.lib)
if "+cuda" in self.spec:
# currently Trilinos doesn't perform the memory fence so
# it relies on blocking CUDA kernel launch.
if self.spec.satisfies("@:13.1.0 +cuda"):
# older releases of Trilinos doesn't perform the memory fence so
# it relies on blocking CUDA kernel launch. This is needed
# in case the dependent app also run a CUDA backend via Trilinos
env.set("CUDA_LAUNCH_BLOCKING", "1")