github
/
ColossalAI
mirror of https://github.com/hpcaitech/ColossalAI
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

[profiler] Fixed bugs in CommProfiler and PcieProfiler (#377)

pull/394/head
HELSON 2022-03-10 17:54:55 +08:00 committed by Frank Lee
parent b5f43acee3
commit 8c18eb0998
2 changed files with 5 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
colossalai/utils/profiler/comm_profiler.py
View File

@ -140,7 +140,7 @@ class CommProfiler(BaseProfiler):
kernel_name, code_location, vol = self.pending_metadata kernel_name, code_location, vol = self.pending_metadata
self.profiler.__exit__(None, None, None) self.profiler.__exit__(None, None, None)
if self.profiler.enabled: if self.profiler.enabled and dist.get_world_size(group) > 1:
assert_flag = 0 assert_flag = 0
current_comm_event = None current_comm_event = None
events = self.profiler.function_events events = self.profiler.function_events

6
colossalai/utils/profiler/pcie_profiler.py
View File

@ -80,8 +80,10 @@ class PcieProfiler(BaseProfiler):
events = self.profiler.function_events events = self.profiler.function_events
for event in events: for event in events:
if event.name == "aten::_to_copy": if event.name == "aten::_to_copy":
current_comm_event = PcieEvent(1, self.data_size * _get_numel(event.input_shapes[0]), t_shape = event.input_shapes[0]
event.cuda_time_total) if len(t_shape) == 0 or event.cuda_time_total == 0:
continue
current_comm_event = PcieEvent(1, self.data_size * _get_numel(t_shape), event.cuda_time_total)
self.total_count += current_comm_event.count self.total_count += current_comm_event.count
self.total_pcie_vol += current_comm_event.pcie_vol self.total_pcie_vol += current_comm_event.pcie_vol
self.total_cuda_time += current_comm_event.cuda_time self.total_cuda_time += current_comm_event.cuda_time