Skip to content

memory_planning algos take the specs as inputs instead of calculating them themselves #9952

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Apr 10, 2025
Merged

memory_planning algos take the specs as inputs instead of calculating them themselves #9952

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 9 additions & 13 deletions backends/cadence/aot/memory_planning.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,7 @@
import math
import typing
from functools import partial
from typing import Iterable, List, Optional, Tuple
from typing import Iterable, List, Optional, Set, Tuple

import torch
from executorch.backends.cadence.aot.memory_constraints import (
Expand Down Expand Up @@ -73,11 +73,11 @@ def collect_specs_from_graph_module(
# the fastest memory available
# flake8: noqa 'position_based_greedy_with_hierarchy' is too complex (13)
def position_based_greedy_with_hierarchy(
graph_module: torch.fx.GraphModule,
alignment: int,
specs: Set[TensorSpec],
graph_module: torch.fx.GraphModule,
graph_signature: ExportGraphSignature,
alloc_graph_input: bool,
alloc_graph_output: bool,
extra_padding: int = 0,
*,
memory_config: MemoryConfig,
mem_constraints: MemConstraints,
Expand Down Expand Up @@ -119,9 +119,7 @@ def memory_available(spec: TensorSpec) -> bool:

# Iterate over all the specs in sorted order
for spec in sorted(
collect_specs_from_graph_module(
graph_module, graph_signature, alloc_graph_input, alloc_graph_output
),
specs,
key=lambda spec: spec.allocated_memory,
reverse=True,
):
Expand Down Expand Up @@ -167,11 +165,11 @@ def memory_available(spec: TensorSpec) -> bool:

# Greedy tensor placement with the heuristics from arxiv.org/pdf/2001.03288.pdf
def greedy_by_size_for_offset_calculation_with_hierarchy(
graph_module: torch.fx.GraphModule,
alignment: int,
specs: Set[TensorSpec],
graph_module: torch.fx.GraphModule,
graph_signature: ExportGraphSignature,
alloc_graph_input: bool,
alloc_graph_output: bool,
extra_padding: int = 0,
*,
memory_config: MemoryConfig,
mem_constraints: MemConstraints,
Expand Down Expand Up @@ -199,9 +197,7 @@ def greedy_by_size_for_offset_calculation_with_hierarchy(

# Iterate over all the specs in sorted order
for spec in sorted(
collect_specs_from_graph_module(
graph_module, graph_signature, alloc_graph_input, alloc_graph_output
),
specs,
key=lambda spec: spec.allocated_memory,
reverse=True,
):
Expand Down
6 changes: 3 additions & 3 deletions backends/vulkan/vulkan_preprocess.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,7 @@
)
from executorch.exir.backend.utils import DelegateMappingBuilder

from executorch.exir.memory_planning import greedy, memory_planning_algorithm_suite
from executorch.exir.memory_planning import greedy, MemoryPlanningAlgorithmSuite
from executorch.exir.pass_base import ExportPass, PassBase

from executorch.exir.passes import MemoryPlanningPass, SpecPropPass
Expand Down Expand Up @@ -199,8 +199,8 @@ def preprocess( # noqa: C901
# Finally, apply dynamic shape passes and memory planning pass. These passes
# must be applied only when the graph structure is finalized.
greedy_memory_planning = partial(greedy, allow_overlapping_allocations=False)
mem_planning_suite = partial(
memory_planning_algorithm_suite, algo_list=[greedy_memory_planning]
mem_planning_suite = MemoryPlanningAlgorithmSuite(
algo_list=[greedy_memory_planning]
)
program = apply_passes(
program,
Expand Down
Loading
Loading