p2p: add CXI transport endpoint

[fergusfinn]

This PR adds a CXI/libfabric transport for the UCCL P2P endpoint used by NIXL. It is separate from the UCCL-EP CXI work in uccl-project#956 and uccl-project#997, but it targets the same Slingshot/Cassini cxi provider.

Benchmarking/testing

export UCCL_P2P_TRANSPORT=cxi
export UCCL_P2P_DISABLE_IPC=1
export UCCL_LIBFABRIC_SO=/opt/libfabric/lib/libfabric.so.1

# Run once per num_iovs in 1,2,4,8,16,32,64,128.
torchrun --nnodes=2 --nproc_per_node=1 --node-rank=<0|1> \
  --master_addr=127.0.0.1 --master_port=<port> \
  p2p/benchmarks/benchmark_uccl.py \
  --local-gpu-idx=<0|1> --device=gpu --sizes=1179648 --iters=16 \
  --num-iovs=<num_iovs> --mode=write --float-type=none

num_iovs	total_bytes	GB/s
1	1179648	19.60
2	2359296	21.58
4	4718592	22.56
8	9437184	23.34
16	18874368	23.81
32	37748736	24.00
64	75497472	24.10
128	150994944	24.17

Then, KVBench in NIXL (currently, kvbench doesn't have UCCL wired in, but it has a plan mode, and nixlbench does have uccl wired in):

 python3 main.py plan \
    --model ./examples/model_deepseek_r1.yaml \
    --model_config ./examples/block-tp1-pp8.yaml \
    --backend UCX \
    --source gpu \
    --destination gpu \
    --runtime_type ASIO \
    --num_iter 16 \
    --warmup_iter 16 \
    --num_threads 1 \
    --page_size 256 \
    --format json
    
  nixlbench \
    --runtime_type ASIO \
    --backend UCCL \
    --worker_type nixl \
    --initiator_seg_type VRAM \
    --target_seg_type VRAM \
    --scheme pairwise \
    --mode SG \
    --op_type WRITE \
    --total_buffer_size 351535104 \
    --num_initiator_dev 1 \
    --num_target_dev 1 \
    --start_block_size 1179648 \
    --max_block_size 1179648 \
    --start_batch_size 298 \
    --max_batch_size 298 \
    --num_iter 16 \
    --warmup_iter 16 \
    --large_blk_iter_ftr 16 \
    --num_threads 1

access	page_size	block_size	batch_size	total_bytes	GB/s	avg_lat_us	avg_prep_us	avg_post_us	avg_tx_us
block	256	1179648	298	351535104	23.141011	51.0	12.0	9.0	15165.0

Also wired it in to vLLM, and vLLM disaggregated prefill completes for Qwen3-0.6b with direct NIXL telemetry showing expected KV transfer volume.

[Copilot]

Pull request overview

This PR adds an optional HPE Slingshot/CXI (libfabric) transport backend to the UCCL P2P engine, selectable at runtime via UCCL_P2P_TRANSPORT=cxi, with build-time enablement via USE_CXI=1.

Changes:

• Introduces CxiEndpoint (libfabric-based) and wires it into the engine’s runtime endpoint variant and notification path.
• Adds a libfabric dlopen/dlsym wrapper scaffold and Makefile build toggles for CXI.
• Adds CXI-related runtime configuration (README) and adjusts one-sided in-flight limits / transient-post handling.

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
p2p/util/transport_type.h	Adds CXI transport selection and helper is_cxi_transport().
p2p/util/common.h	Adds UCCL_POST_TRANSIENT return code used for transient post failures.
p2p/README.md	Documents how to build/run with CXI and new CXI-related env vars.
p2p/Makefile	Adds USE_CXI toggle, libfabric include path, and CXI sources.
p2p/engine.h	Extends GenericEndpoint + MR handle to support CXI endpoint/MR.
p2p/engine.cc	Adds CXI endpoint construction, transient rc handling, and CXI inflight tuning.
p2p/endpoint_wrapper.h	Adds CXI dispatch for regmr/deregmr/read/write and FIFO metadata encoding.
p2p/cxi/fabric_dl.h	New libfabric dlopen/dlsym helper.
p2p/cxi/fabric_dl.cc	New exported fi_* wrappers (currently partial).
p2p/cxi/cxi_endpoint.h	Declares CxiEndpoint and CXI FIFO metadata helpers.
p2p/cxi/cxi_endpoint.cc	Implements the CXI/libfabric endpoint, OOB handshake, MR reg, and RMA ops.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 11 out of 11 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 11 out of 11 changed files in this pull request and generated 2 comments.