[PERF] Symmetric memory allreduce

docs/design/v1/multiprocessing.md

@@ -77,7 +77,7 @@ The `multiproc_xpu_executor` forces the use of `spawn`.
 
 There are other miscellaneous places hard-coding the use of `spawn`:
 
-- <https://github.com/vllm-project/vllm/blob/d05f88679bedd73939251a17c3d785a354b2946c/vllm/distributed/device_communicators/custom_all_reduce_utils.py#L135>
+- <https://github.com/vllm-project/vllm/blob/d05f88679bedd73939251a17c3d785a354b2946c/vllm/distributed/device_communicators/all_reduce_utils.py#L135>
 - <https://github.com/vllm-project/vllm/blob/d05f88679bedd73939251a17c3d785a354b2946c/vllm/entrypoints/openai/api_server.py#L184>
 
 Related PRs:

tests/distributed/test_symm_mem_allreduce.py

@@ -0,0 +1,104 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import random
+import typing
+
+import pytest
+import ray
+import torch
+import torch.distributed as dist
+
+from vllm.distributed.communication_op import tensor_model_parallel_all_reduce
+from vllm.distributed.device_communicators.cuda_communicator import (
+    CudaCommunicator)
+from vllm.distributed.parallel_state import (get_tensor_model_parallel_group,
+                                             get_tp_group)
+from vllm.platforms import current_platform
+
+from ..utils import (ensure_model_parallel_initialized,
+                     init_test_distributed_environment, multi_process_parallel)
+
+torch.manual_seed(42)
+random.seed(44)
+
+test_size_elements = 4 * 1024 * 1024
+
+
+@ray.remote(num_gpus=1, max_calls=1)
+def symm_mem_allreduce(
+    monkeypatch: pytest.MonkeyPatch,
+    tp_size,
+    pp_size,
+    rank,
+    distributed_init_port,
+):
+    with monkeypatch.context() as m:
+        m.delenv("CUDA_VISIBLE_DEVICES", raising=False)
+        device = torch.device(f"cuda:{rank}")
+        torch.cuda.set_device(device)
+
+        init_test_distributed_environment(tp_size, pp_size, rank,
+                                          distributed_init_port)
+        ensure_model_parallel_initialized(tp_size, pp_size)
+
+        dtype = torch.bfloat16
+
+        cuda_communicator = typing.cast(CudaCommunicator,
+                                        get_tp_group().device_communicator)
+        symm_mem_comm = cuda_communicator.symm_mem_comm
+        if symm_mem_comm is None or symm_mem_comm.disabled:
+            pytest.skip("SymmMemCommunicator is not available or disabled.")
+
+        inp_direct_symm_mem = torch.randint(1,
+                                            23, (test_size_elements, ),
+                                            dtype=dtype,
+                                            device=device)
+        if not symm_mem_comm.should_use_symm_mem(inp_direct_symm_mem):
+            pytest.skip(
+                "SymmMemCommunicator isn't used for this world and input size."
+            )
+
+        original_inp_direct_symm_mem = inp_direct_symm_mem.clone()
+        out_direct_symm_mem = symm_mem_comm.all_reduce(inp_direct_symm_mem)
+        assert out_direct_symm_mem is not None
+
+        group = get_tensor_model_parallel_group().device_group
+        dist.all_reduce(original_inp_direct_symm_mem, group=group)
+        torch.testing.assert_close(out_direct_symm_mem,
+                                   original_inp_direct_symm_mem,
+                                   atol=2.5,
+                                   rtol=0.1)
+
+        # Test tensor_model_parallel_all_reduce which should use symm_mem
+        inp_tensor_parallel = torch.randint(-23,
+                                            1, (test_size_elements, ),
+                                            dtype=dtype,
+                                            device=device)
+        original_inp_tensor_parallel = inp_tensor_parallel.clone()
+        out_tensor_parallel = tensor_model_parallel_all_reduce(
+            inp_tensor_parallel)
+        dist.all_reduce(original_inp_tensor_parallel, group=group)
+        torch.testing.assert_close(out_tensor_parallel,
+                                   original_inp_tensor_parallel,
+                                   atol=2.5,
+                                   rtol=0.1)
+
+
+@pytest.mark.skipif(
+    not current_platform.is_cuda(),
+    reason="SymmMemAllreduce is only available for CUDA platforms.")
+@pytest.mark.parametrize("tp_size", [2, 4])
+@pytest.mark.parametrize("pipeline_parallel_size", [1])
+@pytest.mark.parametrize("test_target", [symm_mem_allreduce])
+def test_symm_mem_allreduce(monkeypatch: pytest.MonkeyPatch, tp_size,
+                            pipeline_parallel_size, test_target):
+    world_size = tp_size * pipeline_parallel_size
+    if world_size > torch.cuda.device_count():
+        pytest.skip("Not enough GPUs to run the test.")
+
+    # Enable SymmMemCommunicator
+    monkeypatch.setenv("VLLM_ALLREDUCE_USE_SYMM_MEM", "1")
+
+    multi_process_parallel(monkeypatch, tp_size, pipeline_parallel_size,
+                           test_target)

tools/check_pickle_imports.py

@@ -38,7 +38,7 @@
     'vllm/distributed/utils.py',
     'vllm/distributed/parallel_state.py',
     'vllm/engine/multiprocessing/client.py',
-    'vllm/distributed/device_communicators/custom_all_reduce_utils.py',
+    'vllm/distributed/device_communicators/all_reduce_utils.py',
     'vllm/distributed/device_communicators/shm_broadcast.py',
     'vllm/engine/multiprocessing/engine.py',
     'benchmarks/kernels/graph_machete_bench.py',

vllm/distributed/device_communicators/custom_all_reduce_utils.py → vllm/distributed/device_communicators/all_reduce_utils.py

@@ -23,6 +23,40 @@
 
 logger = init_logger(__name__)
 
+MiB = 1024 * 1024
+
+# Max size for each world size in case symmetric memory is available
+# For different SM architectures
+CUSTOM_ALL_REDUCE_MAX_SIZES = {
+    "9.0": {
+        2: 64 * MiB,  # 64 MB
+        4: 1 * MiB,  # 1 MB
+        6: MiB // 2,  # 512 KB
+        8: MiB // 4,  # 256 KB
+    },
+    "10.0": {
+        2: 2 * MiB,  # 2 MB
+        4: 2 * MiB,  # 2 MB
+        6: 2 * MiB,  # 2 MB
+        8: MiB,  # 1 MB
+    }
+}
+
+SYMM_MEM_ALL_REDUCE_MAX_SIZES = {
+    "9.0": {
+        2: 64 * MiB,  # 64 MB
+        4: 32 * MiB,  # 32 MB
+        6: 64 * MiB,  # 64 MB
+        8: 64 * MiB,  # 64 MB
+    },
+    "10.0": {
+        2: 8 * MiB,  # 8 MB
+        4: 32 * MiB,  # 32 MB
+        6: 128 * MiB,  # 128 MB
+        8: 128 * MiB,  # 128 MB
+    }
+}
+
 
 def producer(batch_src: Sequence[int],
              producer_queue,

vllm/distributed/device_communicators/cuda_communicator.py

@@ -44,6 +44,8 @@ def __init__(self,
             PyNcclCommunicator)
         from vllm.distributed.device_communicators.quick_all_reduce import (
             QuickAllReduce)
+        from vllm.distributed.device_communicators.symm_mem import (
+            SymmMemCommunicator)
 
         self.pynccl_comm: Optional[PyNcclCommunicator] = None
         if use_pynccl and self.world_size > 1:
@@ -54,6 +56,7 @@ def __init__(self,
 
         self.ca_comm: Optional[CustomAllreduce] = None
         self.qr_comm: Optional[QuickAllReduce] = None
+        self.symm_mem_comm: Optional[SymmMemCommunicator] = None
         if use_custom_allreduce and self.world_size > 1:
             # Initialize a custom fast all-reduce implementation.
             self.ca_comm = CustomAllreduce(
@@ -69,6 +72,12 @@ def __init__(self,
                 # currently be an MI300 series.
                 self.qr_comm = QuickAllReduce(group=self.cpu_group,
                                               device=self.device)
+        if envs.VLLM_ALLREDUCE_USE_SYMM_MEM and current_platform.is_cuda():
+            self.symm_mem_comm = SymmMemCommunicator(
+                group=self.cpu_group,
+                device=self.device,
+            )
+
         if self.use_all2all:
             all2all_backend = envs.VLLM_ALL2ALL_BACKEND
             if all2all_backend == "naive":
@@ -105,6 +114,12 @@ def all_reduce(self, input_):
             out = ca_comm.custom_all_reduce(input_)
             assert out is not None
             return out
+        symm_mem_comm = self.symm_mem_comm
+        if symm_mem_comm is not None and not symm_mem_comm.disabled and \
+            symm_mem_comm.should_use_symm_mem(input_):
+            out = symm_mem_comm.all_reduce(input_)
+            assert out is not None
+            return out
         pynccl_comm = self.pynccl_comm
         assert pynccl_comm is not None
         out = pynccl_comm.all_reduce(input_)

vllm/distributed/device_communicators/custom_all_reduce.py

@@ -10,8 +10,8 @@
 
 import vllm.envs as envs
 from vllm import _custom_ops as ops
-from vllm.distributed.device_communicators.custom_all_reduce_utils import (
-    gpu_p2p_access_check)
+from vllm.distributed.device_communicators.all_reduce_utils import (
+    CUSTOM_ALL_REDUCE_MAX_SIZES, gpu_p2p_access_check)
 from vllm.distributed.parallel_state import in_the_same_node_as
 from vllm.logger import init_logger
 from vllm.platforms import current_platform
@@ -109,6 +109,13 @@ def __init__(self,
         # now `device` is a `torch.device` object
         assert isinstance(device, torch.device)
         self.device = device
+        device_capability = current_platform.get_device_capability(
+        ).as_version_str()
+        if (current_platform.is_cuda() and envs.VLLM_ALLREDUCE_USE_SYMM_MEM
+                and device_capability in CUSTOM_ALL_REDUCE_MAX_SIZES):
+            max_size = min(
+                CUSTOM_ALL_REDUCE_MAX_SIZES[device_capability][world_size],
+                max_size)
 
         cuda_visible_devices = envs.CUDA_VISIBLE_DEVICES
         if cuda_visible_devices:

vllm/distributed/device_communicators/symm_mem.py

@@ -0,0 +1,115 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+from typing import Optional, Union
+
+import torch
+import torch.distributed as dist
+from torch.distributed import ProcessGroup
+
+from vllm.distributed.device_communicators.all_reduce_utils import (
+    SYMM_MEM_ALL_REDUCE_MAX_SIZES)
+from vllm.logger import init_logger
+from vllm.platforms import current_platform
+
+try:
+    import torch.distributed._symmetric_memory as torch_symm_mem
+
+    symm_mem_available = True
+except ImportError:
+    symm_mem_available = False
+
+logger = init_logger(__name__)
+
+
+class SymmMemCommunicator:
+
+    # World sizes where multi-mem all-reduce performs the best
+    _WORLD_SIZES_MULTIMEM = {
+        "9.0": [4, 6, 8],
+        "10.0": [6, 8],
+    }
+
+    def __init__(self, group: ProcessGroup, device: Union[int, str,
+                                                          torch.device]):
+        self.disabled = True
+
+        if not symm_mem_available:
+            return
+
+        if not current_platform.is_cuda():
+            logger.warning("SymmMemCommunicator: symmetric "
+                           "memory is not available.")
+            return
+        if isinstance(device, int):
+            device = torch.device(f"cuda:{device}")
+        elif isinstance(device, str):
+            device = torch.device(device)
+        torch.cuda.set_device(device)
+        self.dtype = torch.bfloat16
+        self.device = device
+        self.group = group
+        self.world_size = dist.get_world_size(self.group)
+        self.device_capability = current_platform.get_device_capability(
+        ).as_version_str()
+
+        if self.device_capability not in SYMM_MEM_ALL_REDUCE_MAX_SIZES \
+            or self.device_capability not in self._WORLD_SIZES_MULTIMEM:
+            logger.warning(
+                "SymmMemCommunicator: Device capability %s not supported, "
+                "communicator is not available.",
+                self.device_capability,
+            )
+            return
+        if self.world_size not in SYMM_MEM_ALL_REDUCE_MAX_SIZES[
+                self.device_capability]:
+            logger.warning(
+                "SymmMemCommunicator: World size %d not supported, "
+                "communicator is not available.",
+                self.world_size,
+            )
+            return
+        self.max_size = SYMM_MEM_ALL_REDUCE_MAX_SIZES[self.device_capability][
+            self.world_size]
+        self.buffer = torch_symm_mem.empty(
+            self.max_size // self.dtype.itemsize,
+            device=self.device,
+            dtype=self.dtype,
+        )
+        handle = torch_symm_mem.rendezvous(self.buffer, self.group.group_name)
+        if handle.multicast_ptr == 0:
+            logger.warning("SymmMemCommunicator: symmetric memory "
+                           "multicast operations are not supported.")
+            return
+        self.disabled = False
+
+    def should_use_symm_mem(self, inp: torch.Tensor):
+        if self.disabled:
+            return False
+        if inp.dtype != self.dtype:
+            return False
+        inp_size = inp.numel() * inp.element_size()
+        if inp_size % 4 != 0:
+            return False
+        return inp_size < self.max_size
+
+    def all_reduce(
+            self,
+            inp: torch.Tensor,
+            *,
+            out: Optional[torch.Tensor] = None) -> Optional[torch.Tensor]:
+        if not self.should_use_symm_mem(inp):
+            return None
+        if out is None:
+            out = torch.empty_like(inp)
+        self.buffer[:inp.numel()].copy_(inp.view(-1))
+        if self.world_size in self._WORLD_SIZES_MULTIMEM[
+                self.device_capability]:
+            torch.ops.symm_mem.multimem_all_reduce_(self.buffer[:inp.numel()],
+                                                    "sum",
+                                                    self.group.group_name)
+        else:
+            torch.ops.symm_mem.two_shot_all_reduce_(self.buffer[:inp.numel()],
+                                                    "sum",
+                                                    self.group.group_name)
+        out.copy_(self.buffer[:inp.numel()].view(out.shape))
+        return out

vllm/envs.py

@@ -139,6 +139,7 @@
     VLLM_ROCM_QUICK_REDUCE_CAST_BF16_TO_FP16: bool = True
     VLLM_ROCM_QUICK_REDUCE_MAX_SIZE_BYTES_MB: Optional[int] = None
     VLLM_NIXL_ABORT_REQUEST_TIMEOUT: int = 120
+    VLLM_ALLREDUCE_USE_SYMM_MEM: bool = False
 
 
 def get_default_cache_root():
@@ -620,7 +621,7 @@ def get_vllm_port() -> Optional[int]:
      ("1", "true")),
 
     # By default, vLLM will check the peer-to-peer capability itself,
-    # in case of broken drivers. See https://github.com/vllm-project/vllm/blob/a9b15c606fea67a072416ea0ea115261a2756058/vllm/distributed/device_communicators/custom_all_reduce_utils.py#L101-L108 for details. # noqa
+    # in case of broken drivers. See https://github.com/vllm-project/vllm/blob/a9b15c606fea67a072416ea0ea115261a2756058/vllm/distributed/device_communicators/all_reduce_utils.py#L101-L108 for details. # noqa
     # If this env var is set to 1, vLLM will skip the peer-to-peer check,
     # and trust the driver's peer-to-peer capability report.
     "VLLM_SKIP_P2P_CHECK":
@@ -964,6 +965,9 @@ def get_vllm_port() -> Optional[int]:
     # If set to 1, use the TRTLLM Decode Attention backend in flashinfer.
     "VLLM_USE_TRTLLM_DECODE_ATTENTION":
     lambda: os.getenv("VLLM_USE_TRTLLM_DECODE_ATTENTION", None),
+        # Whether to use pytorch symmetric memory for allreduce
+    "VLLM_ALLREDUCE_USE_SYMM_MEM":
+    lambda: bool(int(os.getenv("VLLM_ALLREDUCE_USE_SYMM_MEM", "0"))),
 }
 
 # --8<-- [end:env-vars-definition]