[UIL&PAL] FlagCX P2P Engine optimization (#478)

---------

Co-authored-by: leoda1 <liud33787@gmail.com>
Co-authored-by: MC952-arch <MC952-arch@qq.com>

Author: 3 people

.github/workflows/unittest-p2p.yml

@@ -0,0 +1,59 @@
+name: "Unit Tests: p2p"
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - main
+
+jobs:
+  unittest-p2p:
+    runs-on: [self-hosted, cx-build]
+    container:
+      image: localhost:5000/flagscale:cuda12.8.1-cudnn9.7.1-python3.12-torch2.7.0-time2507111538
+      options: --gpus all --privileged --ipc=host --ulimit memlock=-1 --ulimit stack=67108864
+    steps:
+      - name: Checkout Code
+        uses: actions/checkout@v6
+        with:
+          repository: ${{ github.event.pull_request.head.repo.full_name }}
+          ref: ${{ github.event.pull_request.head.ref }}
+          ssh-strict: true
+          ssh-user: git
+          persist-credentials: true
+          clean: true
+          sparse-checkout-cone-mode: true
+          fetch-tags: false
+          show-progress: true
+          lfs: false
+          submodules: true
+          set-safe-directory: true
+
+      - name: Build Google Test
+        run: |
+          cd /__w/FlagCX/FlagCX/third-party/googletest
+          mkdir -p build
+          cd build
+          cmake ..
+          make -j$(nproc)
+
+      - name: Build FlagCX
+        run: |
+          cd /__w/FlagCX/FlagCX
+          export MPI_HOME=/usr/local/mpi
+          make -j$(nproc) USE_NVIDIA=1
+
+      - name: Build p2p unit tests
+        run: |
+          cd /__w/FlagCX/FlagCX/test/unittest/p2p
+          export MPI_HOME=/usr/local/mpi
+          make -j$(nproc)
+
+      - name: Run p2p unit tests (requires IB hardware)
+        run: |
+          cd /__w/FlagCX/FlagCX/test/unittest/p2p
+          export MPI_HOME=/usr/local/mpi
+          export LD_LIBRARY_PATH=/__w/FlagCX/FlagCX/build/lib:$LD_LIBRARY_PATH
+          make run-unit

flagcx/adaptor/include/flagcx_net_adaptor.h

@@ -25,6 +25,7 @@ typedef enum {
 //         regMr, regMrDmaBuf, deregMr, isend, irecv, iflush, test,
 //         iput, iget, iputSignal, getDevFromName
 //   v2 — adds iputBatch (optional one-sided batch WRITE)
+//   latest — adds optional batch helpers for one-sided transfers
 
 struct flagcxNetAdaptor_v1 {
   // Basic functions
@@ -135,6 +136,17 @@ struct flagcxNetAdaptor_latest {
                               const size_t *sizes, int srcRank, int dstRank,
                               void **srcHandles, void **dstHandles,
                               void **requests, int *posted);
+  // Optional batch completion test — polls CQ once for multiple requests.
+  // If NULL, caller falls back to per-request test().
+  flagcxResult_t (*testBatch)(void **requests, int nRequests, int *doneFlags,
+                              int *doneCount);
+  // Optional one-side batch READ. Success returns one logical request for the
+  // full batch.
+  flagcxResult_t (*igetBatch)(void *sendComm, int count,
+                              const uint64_t *srcOffs, const uint64_t *dstOffs,
+                              const size_t *sizes, int srcRank, int dstRank,
+                              void *const *srcHandles, void *const *dstHandles,
+                              void **request);
 };
 
 #define flagcxNetAdaptor flagcxNetAdaptor_latest