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

allow multiple values for tensile_architecture and expand the gpu list for rocm-4.3.1 (#26745)

Browse Source
This commit is contained in:
Sreenivasa Murthy Kolam 2021-10-17 23:48:07 -07:00 committed by GitHub
parent cab17c4ac3
commit 1156c7d0a9
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 20 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
var/spack/repos/builtin/packages/rocblas/package.py
View File

@ -26,9 +26,11 @@ class Rocblas(CMakePackage):
version('3.7.0', sha256='9425db5f8e8b6f7fb172d09e2a360025b63a4e54414607709efc5acb28819642') version('3.7.0', sha256='9425db5f8e8b6f7fb172d09e2a360025b63a4e54414607709efc5acb28819642')
version('3.5.0', sha256='8560fabef7f13e8d67da997de2295399f6ec595edfd77e452978c140d5f936f0') version('3.5.0', sha256='8560fabef7f13e8d67da997de2295399f6ec595edfd77e452978c140d5f936f0')
tensile_architecture = ('all', 'gfx803', 'gfx900', 'gfx906', 'gfx908') tensile_architecture = ('all', 'gfx803', 'gfx900', 'gfx906:xnack-', 'gfx908:xnack-',
'gfx90a:xnack+', 'gfx90a:xnack-', 'gfx1010', 'gfx1011',
'gfx1012', 'gfx1030')
variant('tensile_architecture', default='all', values=tensile_architecture, multi=False) variant('tensile_architecture', default='all', values=tensile_architecture, multi=True)
depends_on('cmake@3:', type='build') depends_on('cmake@3:', type='build')
@ -80,14 +82,21 @@ class Rocblas(CMakePackage):
def setup_build_environment(self, env): def setup_build_environment(self, env):
env.set('CXX', self.spec['hip'].hipcc) env.set('CXX', self.spec['hip'].hipcc)
def cmake_args(self): def get_gpulist_for_tensile_support(self):
arch = self.spec.variants['tensile_architecture'].value arch = self.spec.variants['tensile_architecture'].value
if self.spec.satisfies('@4.1.0:'): if arch[0] == 'all':
if arch == 'gfx906' or arch == 'gfx908': if self.spec.satisfies('@:4.0.0'):
arch = arch + ':xnack-' arch_value = self.tensile_architecture[1:2] + 'gfx906,gfx908'
elif self.spec.satisfies('@4.1.0:4.2.0'):
arch_value = self.tensile_architecture[1:4]
elif self.spec.satisfies('@4.3.0:'):
arch_value = self.tensile_architecture[1:]
return arch_value
else:
return arch
def cmake_args(self):
tensile = join_path(self.stage.source_path, 'Tensile') tensile = join_path(self.stage.source_path, 'Tensile')
args = [ args = [
self.define('BUILD_CLIENTS_TESTS', 'OFF'), self.define('BUILD_CLIENTS_TESTS', 'OFF'),
self.define('BUILD_CLIENTS_BENCHMARKS', 'OFF'), self.define('BUILD_CLIENTS_BENCHMARKS', 'OFF'),
@ -109,9 +118,11 @@ def cmake_args(self):
# See https://github.com/ROCmSoftwarePlatform/rocBLAS/commit/c1895ba4bb3f4f5947f3818ebd155cf71a27b634 # See https://github.com/ROCmSoftwarePlatform/rocBLAS/commit/c1895ba4bb3f4f5947f3818ebd155cf71a27b634
if self.spec.satisfies('@:4.2.0'): if self.spec.satisfies('@:4.2.0'):
args.append(self.define('Tensile_ARCHITECTURE', arch)) args.append(self.define('Tensile_ARCHITECTURE',
self.get_gpulist_for_tensile_support()))
else: else:
args.append(self.define('AMDGPU_TARGETS', arch)) args.append(self.define('AMDGPU_TARGETS',
self.get_gpulist_for_tensile_support()))
# See https://github.com/ROCmSoftwarePlatform/rocBLAS/issues/1196 # See https://github.com/ROCmSoftwarePlatform/rocBLAS/issues/1196
if self.spec.satisfies('^cmake@3.21.0:3.21.2'): if self.spec.satisfies('^cmake@3.21.0:3.21.2'):