Merge pull request #2229 from nordmoen/clang_openmp_offload

Add support for building Clang with OpenMP offload support
easybuilders · Nov 12, 2020 · aeda1cc · aeda1cc
2 parents 33c0856 + aec7d0b
commit aeda1cc
Showing 1 changed file with 24 additions and 3 deletions.
diff --git a/easybuild/easyblocks/c/clang.py b/easybuild/easyblocks/c/clang.py
@@ -85,6 +85,8 @@ def extra_options():
             'skip_all_tests': [False, "Skip running of tests", CUSTOM],
             # The sanitizer tests often fail on HPC systems due to the 'weird' environment.
             'skip_sanitizer_tests': [True, "Do not run the sanitizer tests", CUSTOM],
+            'default_cuda_capability': [None, "Default CUDA capability specified for clang, e.g. '7.5'", CUSTOM],
+            'cuda_compute_capabilities': [[], "List of CUDA compute capabilities to build with", CUSTOM],
         })
         # disable regular out-of-source build, too simplistic for Clang to work
         extra_vars['separate_build_dir'][0] = False
@@ -282,9 +284,28 @@ def configure_step(self):
         if self.cfg['parallel']:
             self.make_parallel_opts = "-j %s" % self.cfg['parallel']
 
-        # If we don't want to build with CUDA (not in dependencies) trick CMakes FindCUDA module into
-        # not finding it by using the environment variable which is used as-is and later checked
-        # for a falsy value when determining wether CUDA was found
+        # If 'NVPTX' is in the build targets we assume the user would like OpenMP offload support as well
+        if 'NVPTX' in build_targets:
+            # list of CUDA compute capabilities to use can be specifed in two ways (where (2) overrules (1)):
+            # (1) in the easyconfig file, via the custom cuda_compute_capabilities;
+            # (2) in the EasyBuild configuration, via --cuda-compute-capabilities configuration option;
+            ec_cuda_cc = self.cfg['cuda_compute_capabilities']
+            cfg_cuda_cc = build_option('cuda_compute_capabilities')
+            cuda_cc = cfg_cuda_cc or ec_cuda_cc or []
+            if not cuda_cc:
+                raise EasyBuildError("Can't build Clang with CUDA support "
+                                     "without specifying 'cuda-compute-capabilities'")
+            default_cc = self.cfg['default_cuda_capability'] or min(cuda_cc)
+            if not self.cfg['default_cuda_capability']:
+                print_warning("No default CUDA capability defined! "
+                              "Using '%s' taken as minimum from 'cuda_compute_capabilities'" % default_cc)
+            cuda_cc = [cc.replace('.', '') for cc in cuda_cc]
+            default_cc = default_cc.replace('.', '')
+            self.cfg.update('configopts', '-DCLANG_OPENMP_NVPTX_DEFAULT_ARCH=sm_%s' % default_cc)
+            self.cfg.update('configopts', '-DLIBOMPTARGET_NVPTX_COMPUTE_CAPABILITIES=%s' % ','.join(cuda_cc))
+        # If we don't want to build with CUDA (not in dependencies) trick CMakes FindCUDA module into not finding it by
+        # using the environment variable which is used as-is and later checked for a falsy value when determining
+        # whether CUDA was found
         if not get_software_root('CUDA'):
             setvar('CUDA_NVCC_EXECUTABLE', 'IGNORE')