hlrs-software-stack/spack
7
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Limit the number of parallel jobs launched by Tensile (#33692)

Browse source
This commit is contained in:
Zack Galbreath 2022-11-03 17:26:03 -04:00 committed by GitHub
parent b55509ffa8
commit ccc716f617
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 5 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
var/spack/repos/builtin/packages/rocblas/package.py
View file

@ -248,6 +248,11 @@ def cmake_args(self):
args.append(self.define("Tensile_LIBRARY_FORMAT", "msgpack"))
if self.spec.satisfies("@:4.2.0"):
arch_define_name = "Tensile_ARCHITECTURE"
# Restrict the number of jobs Tensile can spawn.
# If we don't specify otherwise, Tensile creates a job per available core,
# and that consumes a lot of system memory.
# https://github.com/ROCmSoftwarePlatform/Tensile/blob/93e10678a0ced7843d9332b80bc17ebf9a166e8e/Tensile/Parallel.py#L38
args.append(self.define("Tensile_CPU_THREADS", min(16, make_jobs)))
# See https://github.com/ROCmSoftwarePlatform/rocBLAS/commit/c1895ba4bb3f4f5947f3818ebd155cf71a27b634
if "auto" not in self.spec.variants["amdgpu_target"]: