Added profiler communication operations

Fixed bug for learning rate scheduler

colossalai/communication/__init__.py

@@ -1,17 +1,26 @@
 from .collective import all_gather, reduce_scatter, all_reduce, broadcast, reduce
-from .p2p import (send_forward, send_forward_recv_forward,
-                  send_backward_recv_forward, send_backward,
-                  send_backward_recv_backward, send_forward_recv_backward,
-                  send_forward_backward_recv_forward_backward, recv_forward,
-                  recv_backward)
+from .p2p import (send_forward, send_forward_recv_forward, send_backward_recv_forward, send_backward,
+                  send_backward_recv_backward, send_forward_recv_backward, send_forward_backward_recv_forward_backward,
+                  recv_forward, recv_backward)
 from .ring import ring_forward
 from .utils import send_tensor_meta, recv_tensor_meta
 
 __all__ = [
-    'all_gather', 'reduce_scatter', 'all_reduce', 'broadcast', 'reduce',
-    'send_forward', 'send_forward_recv_forward',
-    'send_forward_backward_recv_forward_backward', 'send_backward',
-    'send_backward_recv_backward', 'send_backward_recv_forward',
-    'send_forward_recv_backward', 'recv_backward', 'recv_forward',
-    'ring_forward', 'send_tensor_meta', 'recv_tensor_meta',
+    'all_gather',
+    'reduce_scatter',
+    'all_reduce',
+    'broadcast',
+    'reduce',
+    'send_forward',
+    'send_forward_recv_forward',
+    'send_forward_backward_recv_forward_backward',
+    'send_backward',
+    'send_backward_recv_backward',
+    'send_backward_recv_forward',
+    'send_forward_recv_backward',
+    'recv_backward',
+    'recv_forward',
+    'ring_forward',
+    'send_tensor_meta',
+    'recv_tensor_meta',
 ]

colossalai/trainer/hooks/_lr_scheduler_hook.py

@@ -29,6 +29,7 @@ class LRSchedulerHook(MetricHook):
         self.store_lr_in_state = store_lr_in_state
 
     def after_hook_is_attached(self, trainer):
+        self._check_metric_states_initialization(trainer)
         trainer.states['metrics']['train']['LR'] = LearningRateMetric(epoch_only=self.by_epoch,
                                                                       initial_lr=self.lr_scheduler.get_last_lr()[0])
 

colossalai/utils/__init__.py

@@ -1,12 +1,9 @@
 from .activation_checkpoint import checkpoint
 
-from .common import (clip_grad_norm_fp32, conditional_context,
-                     copy_tensor_parallel_attributes, count_zeros_fp32,
-                     free_port, is_dp_rank_0, is_model_parallel_parameter,
-                     is_moe_parallel_parameter, is_no_pp_or_last_stage,
-                     is_tp_rank_0, is_using_ddp, is_using_pp,
-                     is_using_sequence, multi_tensor_applier,
-                     param_is_not_tensor_parallel_duplicate, print_rank_0,
+from .common import (clip_grad_norm_fp32, conditional_context, copy_tensor_parallel_attributes, count_zeros_fp32,
+                     free_port, is_dp_rank_0, is_model_parallel_parameter, is_moe_parallel_parameter,
+                     is_no_pp_or_last_stage, is_tp_rank_0, is_using_ddp, is_using_pp, is_using_sequence,
+                     multi_tensor_applier, param_is_not_tensor_parallel_duplicate, print_rank_0,
                      switch_virtual_pipeline_parallel_rank, sync_model_param)
 from .cuda import empty_cache, get_current_device, set_to_cuda, synchronize
 from .data_sampler import DataParallelSampler, get_dataloader

colossalai/utils/profiler/__init__.py

@@ -0,0 +1 @@
+from .comm_profiler import enable_communication_prof, communication_prof_show

colossalai/utils/profiler/comm_profiler.py

@@ -0,0 +1,302 @@
+import inspect
+import torch
+from torch.autograd.profiler import profile
+import torch.distributed as dist
+from torch.distributed import ReduceOp
+from colossalai.utils import get_current_device
+from typing import List, Optional
+
+
+def _get_code_location(depth: int):
+    ret = ""
+    length = len(inspect.stack())
+    for i in range(3, min(length, depth + 1)):
+        upper_frame = inspect.stack()[i]
+        function_name = inspect.stack()[i - 1].function
+        info = upper_frame.filename + "(" + str(upper_frame.lineno) + "): " + function_name + "\n"
+        ret += info
+
+    return ret
+
+
+# copied from high version pytorch to support low version
+def _format_time(time_us):
+    """Defines how to format time in FunctionEvent"""
+    US_IN_SECOND = 1000.0 * 1000.0
+    US_IN_MS = 1000.0
+    if time_us >= US_IN_SECOND:
+        return '{:.3f}s'.format(time_us / US_IN_SECOND)
+    if time_us >= US_IN_MS:
+        return '{:.3f}ms'.format(time_us / US_IN_MS)
+    return '{:.3f}us'.format(time_us)
+
+
+# copied from high version pytorch to support low version
+def _format_memory(nbytes):
+    """Returns a formatted memory size string"""
+    KB = 1024
+    MB = 1024 * KB
+    GB = 1024 * MB
+    if (abs(nbytes) >= GB):
+        return '{:.2f} Gb'.format(nbytes * 1.0 / GB)
+    elif (abs(nbytes) >= MB):
+        return '{:.2f} Mb'.format(nbytes * 1.0 / MB)
+    elif (abs(nbytes) >= KB):
+        return '{:.2f} Kb'.format(nbytes * 1.0 / KB)
+    else:
+        return str(nbytes) + ' b'
+
+
+def _format_bandwith(volme: float, time_us: int):
+    sec_div_mb = (1000.0 / 1024.0)**2
+    mb_per_sec = volme / time_us * sec_div_mb
+
+    if mb_per_sec >= 1024.0:
+        return '{:.3f} Gb/s'.format(mb_per_sec / 1024.0)
+    else:
+        return '{:.3f} Mb/s'.format(mb_per_sec)
+
+
+class CommEvent(object):
+    """Communication Event. Used for communication time and communication
+    volume recording.
+    """
+
+    def __init__(self, count: int = 0, comm_vol: float = 0., cuda_time: int = 0):
+        self.self_count = count
+        self.self_comm_vol = comm_vol
+        self.self_cuda_time = cuda_time
+
+    def add(self, rhs):
+        self.self_count += rhs.self_count
+        self.self_comm_vol += rhs.self_comm_vol
+        self.self_cuda_time += rhs.self_cuda_time
+
+
+class CommProfiler(object):
+    """Communication profiler. Records all communication events.
+    """
+
+    def __init__(self, total_count: int = 0, total_comm_vol: float = 0, total_cuda_time: int = 0, prof_depth: int = 3):
+        super().__init__()
+        self.total_count = total_count
+        self.total_comm_vol = total_comm_vol
+        self.total_cuda_time = total_cuda_time
+        self.depth = prof_depth
+
+        self.ops_record = dict()
+        self.profiler = None
+        self.pending_op = None
+        self.pending_metadata = None
+        self.warn_flag = False
+
+    def reset(self):
+        self.total_count = 0
+        self.total_comm_vol = 0
+        self.total_cuda_time = 0
+
+        self.ops_record = dict()
+        self.profiler = None
+        self.pending_op = None
+        self.pending_metadata = None
+        self.warn_flag = False
+
+    def show(self):
+        if self.warn_flag:
+            print("Warnning: there exists multiple communication operations in the same time.\n"
+                  "As a result, the profiling result is not accurate.")
+        print("Collective communication profiling result:",
+              "total cuda time: {}".format(_format_time(self.total_cuda_time)),
+              "average bandwith: {}".format(_format_bandwith(self.total_comm_vol, self.total_cuda_time)),
+              "total number of calls: {}".format(self.total_count),
+              "All events:",
+              sep='\n')
+
+        show_list = sorted(self.ops_record.items(), key=lambda kv: -kv[1].self_cuda_time)
+        for location, event in show_list:
+            print(location,
+                  "self cuda time: {}".format(_format_time(event.self_cuda_time)),
+                  "{:.1f}% of total communication time".format(event.self_cuda_time / self.total_cuda_time * 100.0),
+                  "self communication volme: {}".format(_format_memory(event.self_comm_vol)),
+                  "average bandwith: {}".format(_format_bandwith(event.self_comm_vol, event.self_cuda_time)),
+                  "number of calls: {}".format(event.self_count),
+                  "--------------------",
+                  sep='\n')
+
+    @property
+    def has_aync_op(self):
+        return self.pending_op is not None
+
+    def activate_profiler(self, kn: str, vol: float):
+        self.pending_metadata = (kn, _get_code_location(self.depth), vol)
+        self.profiler = profile(enabled=True, use_cuda=True, use_cpu=True, use_kineto=True)
+        self.profiler.__enter__()
+
+    def close_profiler(self, group=None):
+        assert self.profiler is not None, "There is no running dist op"
+        kernel_name, code_location, vol = self.pending_metadata
+        self.profiler.__exit__(None, None, None)
+
+        if self.profiler.enabled:
+            assert_flag = 0
+            current_comm_event = None
+            events = self.profiler.function_events
+            for event in events:
+                if kernel_name in event.name:
+                    assert assert_flag == 0, "Multiple dist ops has been called "
+                    current_comm_event = CommEvent(1, vol, event.self_cuda_time_total)
+                    assert_flag += 1
+
+            assert current_comm_event is not None, "dist op has not been found"
+
+            buffer = torch.tensor([current_comm_event.self_cuda_time], device=get_current_device())
+            torch_all_reduce(buffer, op=ReduceOp.MIN, group=group)
+            current_comm_event.self_cuda_time = buffer.item()
+
+            self.total_count += current_comm_event.self_count
+            self.total_comm_vol += current_comm_event.self_comm_vol
+            self.total_cuda_time += current_comm_event.self_cuda_time
+            if code_location in self.ops_record:
+                self.ops_record[code_location].add(current_comm_event)
+            else:
+                self.ops_record[code_location] = current_comm_event
+
+        self.profiler = None
+        self.pending_op = None
+        self.pending_metadata = None
+
+    def wait_async_op(self):
+        if self.pending_op is not None:
+            op = self.pending_op
+            op.wait()
+            self.close_profiler()
+
+
+class CommHandler(object):
+    """Communication handler. A dummy handler to wait aync operations.
+    """
+
+    def __init__(self):
+        super().__init__()
+        self.prof = COL_COMM_PROF
+
+    def wait(self):
+        self.prof.wait_async_op()
+
+
+COL_COMM_PROF = CommProfiler()
+torch_all_reduce = dist.all_reduce
+torch_all_gather = dist.all_gather
+torch_reduce_scatter = dist.reduce_scatter
+torch_broadcast = dist.broadcast
+torch_reduce = dist.reduce
+
+
+def enable_communication_prof(depth: int = 0):
+    COL_COMM_PROF.depth = 3 + depth
+    dist.all_reduce = all_reduce
+    dist.all_gather = all_gather
+    dist.reduce_scatter = reduce_scatter
+    dist.broadcast = broadcast
+    dist.reduce = reduce
+
+
+def communication_prof_show():
+    COL_COMM_PROF.show()
+
+
+def async_check():
+    if COL_COMM_PROF.pending_op is not None:
+        COL_COMM_PROF.warn_flag = True
+        COL_COMM_PROF.wait_async_op()
+
+
+def all_reduce(tensor: torch.Tensor,
+               op: ReduceOp = ReduceOp.SUM,
+               group=None,
+               async_op: bool = False) -> Optional[CommHandler]:
+    async_check()
+
+    comm_size = dist.get_world_size(group)
+    correction = 2 * (comm_size - 1) / comm_size
+    comm_vol = correction * tensor.element_size() * tensor.numel()
+    COL_COMM_PROF.activate_profiler("ncclKernel_AllReduce_", comm_vol)
+    COL_COMM_PROF.pending_op = torch_all_reduce(tensor, op, group, async_op)
+
+    if async_op:
+        return CommHandler()
+
+    COL_COMM_PROF.close_profiler(group)
+
+
+def reduce_scatter(output: torch.Tensor,
+                   input_list: List[torch.Tensor],
+                   op: ReduceOp = ReduceOp.SUM,
+                   group=None,
+                   async_op: bool = False) -> Optional[CommHandler]:
+    async_check()
+
+    comm_size = dist.get_world_size(group)
+    correction = (comm_size - 1) / comm_size
+    comm_vol = 0
+    for tensor in input_list:
+        comm_vol += tensor.element_size() * tensor.numel()
+    comm_vol *= correction
+    COL_COMM_PROF.activate_profiler("ncclKernel_ReduceScatter_", comm_vol)
+    COL_COMM_PROF.pending_op = torch_reduce_scatter(output, input_list, op, group, async_op)
+
+    if async_op:
+        return CommHandler()
+
+    COL_COMM_PROF.close_profiler(group)
+
+
+def all_gather(tensor_list: List[torch.Tensor],
+               tensor: torch.Tensor,
+               group=None,
+               async_op: bool = False) -> Optional[CommHandler]:
+    async_check()
+
+    comm_size = dist.get_world_size(group)
+    correction = (comm_size - 1) / comm_size
+    comm_vol = 0
+    for ten in tensor_list:
+        comm_vol += ten.element_size() * ten.numel()
+    comm_vol *= correction
+    COL_COMM_PROF.activate_profiler("ncclKernel_AllGather_", comm_vol)
+    COL_COMM_PROF.pending_op = torch_all_gather(tensor_list, tensor, group, async_op)
+
+    if async_op:
+        return CommHandler()
+
+    COL_COMM_PROF.close_profiler(group)
+
+
+def broadcast(tensor: torch.Tensor, src: int, group=None, async_op: bool = False) -> Optional[CommHandler]:
+    async_check()
+
+    comm_vol = 1.0 * tensor.element_size() * tensor.numel()
+    COL_COMM_PROF.activate_profiler("ncclKernel_Broadcast_", comm_vol)
+    COL_COMM_PROF.pending_op = torch_broadcast(tensor, src, group, async_op)
+
+    if async_op:
+        return CommHandler()
+
+    COL_COMM_PROF.close_profiler(group)
+
+
+def reduce(tensor: torch.Tensor,
+           dst: int,
+           op: ReduceOp = ReduceOp.SUM,
+           group=None,
+           async_op: bool = False) -> Optional[CommHandler]:
+    async_check()
+
+    comm_vol = 1.0 * tensor.element_size() * tensor.numel()
+    COL_COMM_PROF.activate_profiler("ncclKernel_Reduce_", comm_vol)
+    COL_COMM_PROF.pending_op = torch_reduce(tensor, dst, op, group, async_op)
+
+    if async_op:
+        return CommHandler()
+
+    COL_COMM_PROF.close_profiler(group)

tests/test_profiler/test_comm_prof.py

@@ -0,0 +1,40 @@
+from functools import partial
+import torch
+import torch.multiprocessing as mp
+import torch.distributed as dist
+import colossalai
+from colossalai.utils import free_port, get_current_device
+from colossalai.utils.profiler import enable_communication_prof, communication_prof_show
+
+BATCH_SIZE = 1024
+D_MODEL = 1024
+CONFIG = dict(parallel=dict(tensor=dict(mode='1d', size=4)))
+
+
+def run_test(rank, world_size, port):
+    colossalai.launch(config=CONFIG, rank=rank, world_size=world_size, host='localhost', port=port, backend='nccl')
+    inputs = torch.randn(BATCH_SIZE, D_MODEL, dtype=torch.float32, device=get_current_device())
+    outputs = torch.empty(world_size, BATCH_SIZE, D_MODEL, dtype=torch.float32, device=get_current_device())
+    outputs_list = list(torch.chunk(outputs, chunks=world_size, dim=0))
+
+    enable_communication_prof()
+
+    op = dist.all_reduce(inputs, async_op=True)
+    dist.all_gather(outputs_list, inputs)
+    op.wait()
+    dist.reduce_scatter(inputs, outputs_list)
+    dist.broadcast(inputs, 0)
+    dist.reduce(inputs, 0)
+
+    if rank == 0:
+        communication_prof_show()
+
+
+def test_cc_prof():
+    world_size = 4
+    run_func = partial(run_test, world_size=world_size, port=free_port())
+    mp.spawn(run_func, nprocs=world_size)
+
+
+if __name__ == '__main__':
+    test_cc_prof()