GB200 support: SendRecv DSL collective and per-channel executor connections

[Binyang2014]

Summary

GB200 support work: introduces point-to-point send/receive in the MSCCL++ DSL
and extends the executor for split-NVL-domain topologies where some ranks are
NVL-connected within a node and other ranks must communicate across the network.

DSL

• New SendRecv collective with separate input/output buffers
  (python/mscclpp/language/collectives.py).
• New multi-node sendrecv DSL example
  (python/mscclpp/language/tests/multi_node/send_recv.py) with --split_mask
  (group size − 1) and --instances CLI options. Documents the channel-ordering
  trick that keeps signal tags cross-matched between paired peers when
  prev == next.
• BaseBuffer.__getitem__ now accepts slices with None start/stop
  (e.g., buf[:]).

Executor

• One connection (unique QP) per channel entry instead of one per peer.
  Required for HostNoAtomic IB mode where each QP can forward signals to a
  single semaphore. Uses per-peer tag counters so paired ranks agree on tag
  ordering regardless of the order peers appear in each rank's connected_to
  list.
• MEMORY channels now unconditionally use Transport::CudaIpc; only PORT
  channels can use IB. Matches the invariant already enforced by
  getTransportFlags.
• ExecutionContext::connections is now a vector<Connection> indexed by
  channel order (was unordered_map<int, Connection> keyed by peer). Removes
  redundant semaphore fields from ExecutionContext.
• TODO: explicit NVL-domain check in useIB

[Copilot]

Pull request overview

Adds/extends support for a multi-node “sendrecv” communication pattern and adjusts executor connection construction to create distinct connections per channel entry, plus updates the Python executor test harness and verifier kernels to support split-group verification and double-buffering.

Changes:

• Build one Connection (QP) per channel entry with deterministic per-peer tag assignment in the executor.
• Extend Python executor tests/verifier kernels to support sendrecv, optional split_mask grouping, and double-buffer execution.
• Add a new DSL SendRecv collective and a corresponding multi-node test program.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
src/core/include/execution_kernel.hpp	Fixes putWithSignalAndFlush argument ordering/selection in device-side put handling.
src/core/executor/executor.cc	Switches to per-channel connections (vector-based) and uses per-peer tag counters for stable matching.
python/test/executor_test.py	Adds sendrecv/double-buffer and split-mask support to benchmarking and correctness harness.
python/test/executor_test_verifier.cu	Updates fill/test kernels to accept split_mask and adds sendrecv verification kernels.
python/mscclpp/language/tests/multi_node/send_recv.py	New multi-node DSL test program for sendrecv with split grouping.
python/mscclpp/language/rank.py	Makes buffer slicing more robust by handling slice(None, None) and non-slice indexing.
python/mscclpp/language/collectives.py	Introduces SendRecv collective and buffer initialization for it.