DLcomm 0.3.2__tar.gz → 0.3.4__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {dlcomm-0.3.2 → dlcomm-0.3.4}/PKG-INFO +2 -3
- dlcomm-0.3.4/dl_comm/__init__.py +1 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/comm/comm_setup.py +19 -2
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/dl_comm_main.py +1 -1
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/utils/utility.py +1 -1
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dlcomm.egg-info/PKG-INFO +2 -3
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dlcomm.egg-info/requires.txt +1 -3
- {dlcomm-0.3.2 → dlcomm-0.3.4}/pyproject.toml +2 -4
- dlcomm-0.3.2/dl_comm/__init__.py +0 -1
- {dlcomm-0.3.2 → dlcomm-0.3.4}/README.md +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/analysis/__init__.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/analysis/bandwidth.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/analysis/ccl_parser.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/analysis/correctness.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/comm/__init__.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/comm/collectives.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/config/__init__.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/config/config_spec.json +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/config/mpi_utils.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/config/setup_algorithms.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/config/system_info.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/config/validation.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/timer/__init__.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/timer/timer.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dl_comm/utils/__init__.py +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dlcomm.egg-info/SOURCES.txt +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dlcomm.egg-info/dependency_links.txt +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dlcomm.egg-info/entry_points.txt +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/dlcomm.egg-info/top_level.txt +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/setup.cfg +0 -0
- {dlcomm-0.3.2 → dlcomm-0.3.4}/tests/test_mxm_unitrace.py +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: dlcomm
|
|
3
|
-
Version: 0.3.
|
|
3
|
+
Version: 0.3.4
|
|
4
4
|
Summary: Distributed GPU Communication Benchmarking Framework for Deep Learning
|
|
5
5
|
Author-email: Musa Cim <musaoguzhancim@gmail.com>, Kaushik Velusamy <kaushikvelusamy@gmail.com>
|
|
6
6
|
Project-URL: Homepage, https://github.com/argonne-lcf/DLcomm_benchmark
|
|
@@ -27,8 +27,7 @@ Requires-Dist: mpi4py>=3.1.0
|
|
|
27
27
|
Requires-Dist: hydra-core>=1.1.0
|
|
28
28
|
Requires-Dist: omegaconf>=2.1.0
|
|
29
29
|
Requires-Dist: pytz>=2021.1
|
|
30
|
-
|
|
31
|
-
Requires-Dist: torch>=1.12.0; extra == "pytorch"
|
|
30
|
+
Requires-Dist: torch>=1.12.0
|
|
32
31
|
Provides-Extra: intel
|
|
33
32
|
Requires-Dist: intel-extension-for-pytorch; extra == "intel"
|
|
34
33
|
Requires-Dist: oneccl-bindings-for-pytorch; extra == "intel"
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
__version__ = "0.3.4"
|
|
@@ -2,11 +2,15 @@
|
|
|
2
2
|
from mpi4py import MPI
|
|
3
3
|
from omegaconf import DictConfig
|
|
4
4
|
from dl_comm.timer import timer
|
|
5
|
+
import os
|
|
5
6
|
|
|
6
7
|
|
|
7
8
|
def allocate_device(device_type, assigned_device_id, log, mpi_rank, framework):
|
|
8
|
-
|
|
9
|
+
|
|
10
|
+
# FRAMEWORK-PYTORCH
|
|
9
11
|
if framework == 'pytorch':
|
|
12
|
+
|
|
13
|
+
# GPU ALLOCATION
|
|
10
14
|
if device_type == 'gpu':
|
|
11
15
|
if torch.cuda.is_available():
|
|
12
16
|
device = torch.device(f"cuda:{assigned_device_id}")
|
|
@@ -15,9 +19,22 @@ def allocate_device(device_type, assigned_device_id, log, mpi_rank, framework):
|
|
|
15
19
|
elif torch.xpu.is_available():
|
|
16
20
|
device = torch.device(f"xpu:{assigned_device_id}")
|
|
17
21
|
return device
|
|
22
|
+
|
|
23
|
+
|
|
24
|
+
# CPU ALLOCATION
|
|
18
25
|
elif device_type == 'cpu':
|
|
26
|
+
|
|
27
|
+
cpu_binding = os.environ.get("CPU_BINDING")
|
|
28
|
+
if cpu_binding and cpu_binding.startswith("list:"):
|
|
29
|
+
cpu_cores = [int(x) for x in cpu_binding[5:].split(":")]
|
|
30
|
+
cpu_core = cpu_cores[assigned_device_id]
|
|
31
|
+
os.sched_setaffinity(0, {cpu_core})
|
|
32
|
+
|
|
33
|
+
log.info(f"[CPU] Rank {mpi_rank} bound to CPU core {cpu_core}")
|
|
34
|
+
|
|
19
35
|
return torch.device('cpu')
|
|
20
|
-
|
|
36
|
+
|
|
37
|
+
# FRAMEWORK-JAX
|
|
21
38
|
elif framework == 'jax':
|
|
22
39
|
pass
|
|
23
40
|
|
|
@@ -428,8 +428,8 @@ def main(cfg: DictConfig):
|
|
|
428
428
|
log.info("")
|
|
429
429
|
|
|
430
430
|
|
|
431
|
-
|
|
432
431
|
|
|
432
|
+
|
|
433
433
|
# ----------------------------------------------------------------------------
|
|
434
434
|
# COLLECTIVE OP EXECUTION (TIMED)
|
|
435
435
|
# ----------------------------------------------------------------------------
|
|
@@ -120,7 +120,7 @@ def utcnow(format=LOG_TS_FORMAT):
|
|
|
120
120
|
return datetime.now().strftime(format)
|
|
121
121
|
|
|
122
122
|
|
|
123
|
-
def dummy_mxm_compute(device, dtype=None, size=None, framework):
|
|
123
|
+
def dummy_mxm_compute(device, dtype=None, size=None, framework="pytorch"):
|
|
124
124
|
if framework == 'pytorch':
|
|
125
125
|
import torch
|
|
126
126
|
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: dlcomm
|
|
3
|
-
Version: 0.3.
|
|
3
|
+
Version: 0.3.4
|
|
4
4
|
Summary: Distributed GPU Communication Benchmarking Framework for Deep Learning
|
|
5
5
|
Author-email: Musa Cim <musaoguzhancim@gmail.com>, Kaushik Velusamy <kaushikvelusamy@gmail.com>
|
|
6
6
|
Project-URL: Homepage, https://github.com/argonne-lcf/DLcomm_benchmark
|
|
@@ -27,8 +27,7 @@ Requires-Dist: mpi4py>=3.1.0
|
|
|
27
27
|
Requires-Dist: hydra-core>=1.1.0
|
|
28
28
|
Requires-Dist: omegaconf>=2.1.0
|
|
29
29
|
Requires-Dist: pytz>=2021.1
|
|
30
|
-
|
|
31
|
-
Requires-Dist: torch>=1.12.0; extra == "pytorch"
|
|
30
|
+
Requires-Dist: torch>=1.12.0
|
|
32
31
|
Provides-Extra: intel
|
|
33
32
|
Requires-Dist: intel-extension-for-pytorch; extra == "intel"
|
|
34
33
|
Requires-Dist: oneccl-bindings-for-pytorch; extra == "intel"
|
|
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "dlcomm"
|
|
7
|
-
version = "0.3.
|
|
7
|
+
version = "0.3.4"
|
|
8
8
|
description = "Distributed GPU Communication Benchmarking Framework for Deep Learning"
|
|
9
9
|
authors = [
|
|
10
10
|
{ name = "Musa Cim", email = "musaoguzhancim@gmail.com" },
|
|
@@ -34,6 +34,7 @@ dependencies = [
|
|
|
34
34
|
"hydra-core>=1.1.0",
|
|
35
35
|
"omegaconf>=2.1.0",
|
|
36
36
|
"pytz>=2021.1",
|
|
37
|
+
"torch>=1.12.0",
|
|
37
38
|
]
|
|
38
39
|
|
|
39
40
|
[project.urls]
|
|
@@ -46,9 +47,6 @@ Repository = "https://github.com/argonne-lcf/DLcomm_benchmark"
|
|
|
46
47
|
dlcomm = "dl_comm.dl_comm_main:main"
|
|
47
48
|
|
|
48
49
|
[project.optional-dependencies]
|
|
49
|
-
pytorch = [
|
|
50
|
-
"torch>=1.12.0"
|
|
51
|
-
]
|
|
52
50
|
intel = [
|
|
53
51
|
"intel-extension-for-pytorch",
|
|
54
52
|
"oneccl-bindings-for-pytorch"
|
dlcomm-0.3.2/dl_comm/__init__.py
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
__version__ = "0.2.0"
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|