[Feat] Intranode Dispatch&Combine Kernel

.github/workflows/flydsl.yaml

@@ -350,15 +350,15 @@ jobs:
           docker rm flydsl_test
 
   # ---------------------------------------------------------------------------
-  # Multi-GPU allreduce tests: ONLY for 8-GPU runners.
+  # Multi-GPU communication operator tests: ONLY for 8-GPU runners.
   # Runs on BOTH linux-flydsl-mi325-8 AND linux-flydsl-mi355-8 independently.
   # Triggered when PR has label "multi-gpu" (added by a maintainer), or when
   # the workflow is manually dispatched.
   # fail-fast: false ensures both runners always complete even if one fails.
   # ---------------------------------------------------------------------------
   multi-gpu:
     needs: test
-    name: Multi-GPU AllReduce Tests (${{ matrix.runners }})
+    name: Multi-GPU Communication Operator Tests (${{ matrix.runners }})
     timeout-minutes: 120
     env:
       # Keep checkout independent of runner-local git-cache rewrites.
@@ -469,6 +469,18 @@ jobs:
               -m multi_gpu -v --no-header --tb=short
           "
 
+      - name: Run multi-GPU dispatch/combine CI sweep (8-GPU, accuracy + cudagraph perf)
+        timeout-minutes: 30
+        run: |
+          docker exec flydsl_test bash -c "
+            cd /flydsl-test &&
+            python tests/kernels/test_profiler_dispatch_combine.py \
+              --ci-sweep \
+              --world-size 8 \
+              --port 29503 \
+              --output-dir /tmp/flydsl_ci_sweep
+          "
+
       - name: Run multi-GPU allreduce tests
         timeout-minutes: 30
         run: |