Intel/CI: Add OneCCL CPU tests with DSA + shm to the CI

- Add OneCCL DSA stage. - There was an issue with libfabric being built with ze and oneccl built without. This led to failure and CCL transport being switched to MPI because oneccl looked for the library. To avoid this, an else case is added with --with-ze=no. Signed-off-by: Juee Himalbhai Desai <[email protected]>
ofiwg · Mar 21, 2024 · 0295f52 · 0295f52
1 parent 2aa1d90
commit 0295f52
Show file tree

Hide file tree

Showing 2 changed files with 14 additions and 8 deletions.
diff --git a/contrib/intel/jenkins/Jenkinsfile b/contrib/intel/jenkins/Jenkinsfile
@@ -694,18 +694,22 @@ pipeline {
             }
           }
         }
-	stage ('oneCCL') {
+	      stage ('oneCCL') {
           steps {
             script {
               dir (RUN_LOCATION) {
-		run_middleware([["verbs", null]], "oneCCL",
+		            run_middleware([["verbs", null]], "oneCCL",
                                "oneccl", "water", "squirtle,totodile", "2")
-		run_middleware([["shm", null]], "oneCCL",
-			       "oneccl", "grass", "bulbasaur", "1")
-		run_middleware([["psm3", null]], "oneCCL",
-			       "oneccl", "water", "squirtle", "2")
-		run_middleware([["tcp", null]], "oneCCL",
-			       "oneccl", "grass", "bulbasaur", "2")
+		            run_middleware([["shm", null]], "oneCCL",
+			                         "oneccl", "grass", "bulbasaur", "1")
+		            run_middleware([["psm3", null]], "oneCCL",
+			                         "oneccl", "water", "squirtle", "2")
+		            run_middleware([["tcp", null]], "oneCCL",
+			                         "oneccl", "grass", "bulbasaur", "2")
+                run_middleware([["shm", null]], "oneCCL_DSA",
+                               "oneccl", "electric", "pikachu", "1", null, null,
+                               """CCL_ATL_SHM=1 FI_SHM_DISABLE_CMA=1 \
+                               FI_SHM_USE_DSA_SAR=1 FI_LOG_LEVEL=warn""")
               }
             }
           }

diff --git a/contrib/intel/jenkins/build.py b/contrib/intel/jenkins/build.py
@@ -36,6 +36,8 @@ def build_libfabric(libfab_install_path, mode, hw_type, gpu=False, cuda=False):
 
 	if (gpu):
 		config_cmd.append('--enable-ze-dlopen')
+	else:
+		config_cmd.append('--with-ze=no')
 
 	if (cuda):
 		config_cmd.append(f'--with-cuda={os.environ["CUDA_INSTALL"]}')