ColossalAI/colossalai/autochunk/search_chunk.py

301 lines
11 KiB
Python
Raw Normal View History

2023-01-06 06:21:49 +00:00
import copy
2023-01-09 09:11:51 +00:00
from typing import Dict, List, Tuple
2023-01-09 08:54:08 +00:00
from torch.fx.node import Node
2023-01-06 06:21:49 +00:00
2023-01-09 05:46:17 +00:00
from .estimate_memory import EstimateMemory
2023-01-06 09:31:59 +00:00
from .reorder_graph import ReorderGraph
from .select_chunk import SelectChunk
2023-01-06 09:24:23 +00:00
from .trace_flow import TraceFlow
2023-01-09 09:25:13 +00:00
from .trace_indice import TraceIndice
from .utils import NodeMgr, get_logger, is_non_compute_node, is_non_compute_node_except_placeholder
2023-01-06 06:14:45 +00:00
2023-01-06 09:09:37 +00:00
class SearchChunk(object):
2023-01-09 08:54:08 +00:00
"""
This is the core class for AutoChunk.
It defines the framework of the strategy of AutoChunk.
Chunks will be selected one by one until search stops.
2023-01-09 08:54:08 +00:00
The chunk search is as follows:
1. find the peak memory node
2. find the max chunk region according to the peak memory node
3. find all possible chunk regions in the max chunk region
4. find the best chunk region for current status
5. goto 1
Attributes:
gm: graph model
print_mem (bool): print estimated memory
trace_index: trace the flow of every dim of every node to find all free dims
trace_flow: determine the region chunk strategy
reorder_graph: reorder nodes to improve chunk efficiency
estimate_memory: estimate memory with chunk
select_chunk: select the best chunk region
Args:
gm: graph model
max_memory (int): max memory in MB
print_mem (bool): print estimated memory
"""
def __init__(self, gm, max_memory=None, print_mem=False, print_progress=False) -> None:
2023-01-06 06:19:45 +00:00
self.print_mem = print_mem
self.max_memory = max_memory
self.print_progress = print_progress
self.node_mgr = NodeMgr(list(gm.graph.nodes))
self.trace_indice = TraceIndice(self.node_mgr)
self.estimate_memory = EstimateMemory()
self._init_trace()
self.trace_flow = TraceFlow(self.trace_indice, self.node_mgr)
self.reorder_graph = ReorderGraph(self.trace_indice, self.node_mgr)
2023-01-06 09:24:23 +00:00
self.select_chunk = SelectChunk(
2023-01-09 09:25:13 +00:00
self.trace_indice,
2023-01-06 09:31:59 +00:00
self.estimate_memory,
self.reorder_graph,
self.node_mgr,
2023-01-06 09:31:59 +00:00
max_memory=max_memory,
2023-01-06 06:14:45 +00:00
)
def _init_trace(self) -> None:
"""
find the max trace range for every node
reduce the computation complexity of trace_indice
"""
# find all max ranges
active_nodes = self.estimate_memory.estimate_chunk_inference_mem(self.node_mgr.get_node_list())[2]
# set trace range and do the trace
if self.print_progress:
get_logger().info("AutoChunk start tracing indice")
self.trace_indice.set_active_nodes(active_nodes)
self.trace_indice.trace_indice()
def _find_peak_region(self, mem_peak: List) -> int:
"""
find peak node, along with its neighbor nodes exceeds max mem
"""
2023-01-06 06:14:45 +00:00
max_value = max(mem_peak)
max_idx = mem_peak.index(max_value)
peak_region = [max_idx, max_idx]
if self.max_memory is None:
return peak_region
# to left
count = 0
for i in range(max_idx - 1, -1, -1):
if mem_peak[i] > self.max_memory:
peak_region[0] = i
else:
count += 1
if count >= 3:
break
# to right
count = 0
for i in range(max_idx + 1, len(mem_peak) - 1):
if mem_peak[i] > self.max_memory:
peak_region[1] = i
count = 0
else:
count += 1
if count >= 3:
break
2023-01-09 08:54:08 +00:00
return peak_region
2023-01-06 06:14:45 +00:00
def _search_max_chunk_region(self, active_node: List, peak_region: int, chunk_regions: List = None) -> Tuple:
2023-01-09 08:54:08 +00:00
"""
Search max chunk region according to peak memory node
Chunk region starts extending from the peak node, stops where free var num is min
2023-01-06 06:14:45 +00:00
2023-01-09 08:54:08 +00:00
Args:
active_node (List): active node status for every node
peak_node_idx (int): peak memory node idx
2023-01-10 02:44:07 +00:00
chunk_regions (List): chunk region infos
2023-01-09 08:54:08 +00:00
Returns:
chunk_region_start (int)
chunk_region_end (int)
"""
# check if peak node already in chunk info
if chunk_regions is not None:
for i in chunk_regions:
if (
i["region"][0] < peak_region[0] <= i["region"][1]
or i["region"][0] < peak_region[1] <= i["region"][1]
):
return None
2023-01-06 06:14:45 +00:00
active_node_num = [len(i) for i in active_node]
window_size = 100
# search min for start
min_num = 1e4
for i in range(peak_region[0], max(peak_region[0] - window_size, -1), -1):
if active_node_num[i] < min_num:
min_num = active_node_num[i]
chunk_region_start = i
# search min for end
min_num = 1e4
for i in range(peak_region[1], min(peak_region[1] + window_size, len(active_node_num))):
if active_node_num[i] < min_num:
min_num = active_node_num[i]
2023-01-06 06:14:45 +00:00
chunk_region_end = i
# avoid chunk regions overlap
if chunk_regions is not None:
for i in chunk_regions:
region = i["region"]
if chunk_region_start >= region[0] and chunk_region_end <= region[1]:
return None
elif region[0] <= chunk_region_start <= region[1] and chunk_region_end > region[1]:
chunk_region_start = region[1] + 1
elif region[0] <= chunk_region_end <= region[1] and chunk_region_start < region[0]:
chunk_region_end = region[0] - 1
2023-01-06 06:14:45 +00:00
return chunk_region_start, chunk_region_end
def _find_chunk_info(self, input_trace, output_trace, start_idx, end_idx) -> List:
"""
Find chunk info for a region.
We are given the region start and region end, and need to find out all chunk info for it.
We first loop every dim of start node and end node, to see if we can find dim pair,
which is linked in a flow and not computed.
If found, we then search flow in the whole region to find out all chunk infos.
Args:
input_trace (List): node's input trace in region
output_trace (List): node's output trace in region
start_idx (int): region start node index
end_idx (int): region end node index
Returns:
chunk_infos: possible regions found
"""
start_traces = input_trace[start_idx]
if len(start_traces) > 1: # TODO need to be removed
return []
end_trace = output_trace[end_idx]
end_node = self.node_mgr.get_node_by_idx(end_idx)
chunk_infos = []
for end_dim, _ in enumerate(end_trace["indice"]):
for start_node, start_trace in start_traces.items():
for start_dim, _ in enumerate(start_trace["indice"]):
if not self.trace_flow.check_region_start_end(
start_node, start_dim, start_idx, end_node, end_dim, end_idx
):
continue
# flow search
chunk_info = self.trace_flow.flow_search(start_idx, start_dim, end_idx, end_dim)
if chunk_info is None:
continue
chunk_infos.append(chunk_info)
return chunk_infos
def _search_possible_chunk_regions(self, max_chunk_region: Tuple, peak_region: Node) -> List:
2023-01-09 09:11:51 +00:00
"""
Search every possible region within the max chunk region.
Args:
max_chunk_region (Tuple)
peak_node (Node): peak memory node
Returns:
possible_chunk_region (List)
"""
2023-01-06 06:14:45 +00:00
possible_chunk_region = []
2023-01-09 09:34:30 +00:00
output_trace = copy.deepcopy(self.trace_indice.indice_trace_list)
input_trace = [] # trace of a node's input nodes
for _, n in enumerate(self.node_mgr.get_node_list()):
2023-01-06 06:14:45 +00:00
cur_trace = {}
for arg in n.args:
if type(arg) == type(n) and not is_non_compute_node_except_placeholder(arg):
2023-01-09 09:25:13 +00:00
cur_trace[arg] = self.trace_indice._find_trace_from_node(arg)
2023-01-06 06:14:45 +00:00
input_trace.append(cur_trace)
for start_idx in range(max_chunk_region[0], peak_region[0] + 1):
for end_idx in range(peak_region[1], max_chunk_region[1] + 1):
2023-01-06 06:14:45 +00:00
# skip non compute nodes
if is_non_compute_node(self.node_mgr.get_node_by_idx(start_idx)) or is_non_compute_node(
self.node_mgr.get_node_by_idx(end_idx)
):
2023-01-06 06:14:45 +00:00
continue
# select free dim
chunk_info = self._find_chunk_info(input_trace, output_trace, start_idx, end_idx)
2023-01-06 06:14:45 +00:00
if len(chunk_info) > 0:
possible_chunk_region.extend(chunk_info)
return possible_chunk_region
2023-01-09 09:11:51 +00:00
def _step_search(
self,
mem_peak: List[float],
active_node: List[List[Node]],
chunk_infos: List[Dict],
) -> Dict:
"""
Find one chunk region
The chunk search is as follows:
1. find the peak memory node
2. find the max chunk region according to the peak memory node
3. find all possible chunk regions in the max chunk region
4. find the best chunk region for current status
Args:
mem_peak (List): peak memory for every node
active_node (List[List[Node]]): active node for every node
chunk_infos (List[Dict]): all chunk info
Returns:
best_chunk_region (Dict)
"""
peak_region = self._find_peak_region(mem_peak)
max_chunk_region = self._search_max_chunk_region(active_node, peak_region, chunk_infos)
2023-01-06 06:14:45 +00:00
if max_chunk_region == None:
return None
possible_chunk_regions = self._search_possible_chunk_regions(max_chunk_region, peak_region)
best_chunk_region = self.select_chunk._select_best_chunk_region(possible_chunk_regions, chunk_infos, mem_peak)
2023-01-06 08:53:01 +00:00
best_chunk_region = self.reorder_graph.reorder_all(best_chunk_region)
2023-01-06 06:14:45 +00:00
return best_chunk_region
2023-01-09 09:11:51 +00:00
def search_region(self) -> Dict:
"""
Search all chunk regions:
1. Estimate current memory
2. Find best chunk for current memory
3. goto 1
Returns:
chunk_infos (Dict)
"""
if self.print_progress:
get_logger().info("AutoChunk start searching chunk regions")
2023-01-06 06:14:45 +00:00
chunk_infos = []
init_mem_peak, _, active_node = self.estimate_memory.estimate_chunk_inference_mem(self.node_mgr.get_node_list())
2023-01-06 06:14:45 +00:00
mem_peak = init_mem_peak
while True:
chunk_info = self._step_search(mem_peak, active_node, chunk_infos)
if chunk_info is None:
break
chunk_infos.append(chunk_info)
mem_peak, _, active_node = self.estimate_memory.estimate_chunk_inference_mem(
self.node_mgr.get_node_list(), chunk_infos
)
if self.print_progress:
get_logger().info(
"AutoChunk find chunk region %d = (%d, %d)"
% (len(chunk_infos), chunk_info["region"][0], chunk_info["region"][1])
)
2023-01-06 06:19:45 +00:00
if self.print_mem:
self.print_mem = False
self.estimate_memory.estimate_chunk_inference_mem(
self.node_mgr.get_node_list(), chunk_infos, print_mem=True
)
2023-01-06 06:14:45 +00:00
return chunk_infos