contrib/intel/jenkins: Fix bug where device is slurm partition

Slurm partition name is getting passed in as the device when it shouldn't be. Add cuda.exclude file to remove failing tests Skip xd2di. Since there is only one device this test is impossible. Signed-off-by: Zach Dworkin <zachary.dworkin@intel.com>
ofiwg · Mar 15, 2024 · 0377d56 · 0377d56
1 parent ccb0f0a
commit 0377d56
Show file tree

Hide file tree

Showing 5 changed files with 42 additions and 7 deletions.
diff --git a/contrib/intel/jenkins/Jenkinsfile b/contrib/intel/jenkins/Jenkinsfile
@@ -776,8 +776,8 @@ pipeline {
                              "shm", null, null, "h2d")
                 run_fabtests("cuda_v1_shm", "cyndaquil", "cyndaquil", "1",
                              "shm", null, null, "d2d")
-                run_fabtests("cuda_v1_shm", "cyndaquil", "cyndaquil", "1",
-                             "shm", null, null, "xd2d")
+                // run_fabtests("cuda_v1_shm", "cyndaquil", "cyndaquil", "1",
+                //              "shm", null, null, "xd2d")
               }
             }
           }
@@ -790,8 +790,8 @@ pipeline {
                              null, null, "h2d")
                 run_fabtests("cuda_v2_shm", "quilava", "quilava", "1", "shm",
                              null, null, "d2d")
-                run_fabtests("cuda_v2_shm", "quilava ", "quilava", "1", "shm",
-                             null, null, "xd2d")
+                // run_fabtests("cuda_v2_shm", "quilava ", "quilava", "1", "shm",
+                //              null, null, "xd2d")
               }
             }
           }

diff --git a/contrib/intel/jenkins/build.py b/contrib/intel/jenkins/build.py
@@ -56,6 +56,9 @@ def build_fabtests(libfab_install_path, mode, cuda=False):
 		config_cmd = ['./configure', f'--prefix={libfab_install_path}',
 					  f'--with-libfabric={libfab_install_path}']
 
+	if cuda:
+		config_cmd.append(f'--with-cuda={os.environ["CUDA_INSTALL"]}')
+
 	common.run_command(['./autogen.sh'])
 	common.run_command(config_cmd)
 	common.run_command(['make','clean'])

diff --git a/contrib/intel/jenkins/tests.py b/contrib/intel/jenkins/tests.py
@@ -106,13 +106,16 @@ def __init__(self, jobname, buildno, testname, hw, core_prov, fabric,
                          util_prov, way)
         self.fabtestpath = f'{self.libfab_installpath}/bin'
         self.fabtestconfigpath = f'{self.libfab_installpath}/share/fabtests'
+        self.device = cloudbees_config.fabric_map[self.hw]
 
     def get_exclude_file(self):
         path = self.libfab_installpath
         efile_path = f'{path}/share/fabtests/test_configs'
 
         if self.hw == 'ivysaur':
             efile = f'{efile_path}/{self.core_prov}/io_uring.exclude'
+        elif self.hw == 'cyndaquil' or self.hw == 'quilava':
+            efile = f'{efile_path}/{self.core_prov}/cuda.exclude'
         else:
             prov = self.util_prov if self.util_prov else self.core_prov
             efile_old = f'{efile_path}/{prov}/{prov}.exclude'
@@ -155,11 +158,11 @@ def options(self):
             opts += "-t all "
 
         if (self.way == 'h2d'):
-            opts += f"-C \"-H\" -L \"-D {self.hw}\" "
+            opts += f"-C \"-H\" -L \"-D {self.device}\" "
         elif (self.way == 'd2d'):
-            opts += f"-C \"-D {self.hw}\" -L \"-D {self.hw}\" "
+            opts += f"-C \"-D {self.device}\" -L \"-D {self.device}\" "
         elif (self.way == 'xd2d'):
-            opts += f"-C \"-D {self.hw}\" -L \"-D {self.hw} -i 1\" "
+            opts += f"-C \"-D {self.device}\" -L \"-D {self.device} -i 1\" "
 
         if (self.core_prov == 'sockets' and self.ofi_build_mode == 'reg'):
             complex_test_file = f'{self.libfab_installpath}/share/fabtests/'\

diff --git a/fabtests/Makefile.am b/fabtests/Makefile.am
@@ -125,6 +125,7 @@ nobase_dist_config_DATA = \
 	test_configs/ofi_rxd/ofi_rxd.exclude \
 	test_configs/shm/all.test \
 	test_configs/shm/shm.exclude \
+	test_configs/shm/cuda.exclude \
 	test_configs/shm/quick.test \
 	test_configs/shm/verify.test \
 	test_configs/sm2/quick.test \

diff --git a/fabtests/test_configs/shm/cuda.exclude b/fabtests/test_configs/shm/cuda.exclude
@@ -0,0 +1,28 @@
+# Regex patterns of tests to exclude in runfabtests.sh
+
+inject_test
+^fi_msg
+-e msg
+^fi_dgram
+-e dgram
+rdm_tagged_peek
+multi_ep
+av_xfer
+unexpected_msg
+multi_recv
+
+# Exclude tests that use sread/polling
+rdm_cntr_pingpong
+poll
+
+# Exclude tests with unsupported capabilities
+-k
+cm_data
+trigger
+shared_ctx
+scalable_ep
+shared_av
+multi_mr
+av_test
+
+multinode