Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
Menu
Open sidebar
OpenDAS
DeepEP
Commits
c939644c
Unverified
Commit
c939644c
authored
Sep 22, 2025
by
Shangyan Zhou
Committed by
GitHub
Sep 22, 2025
Browse files
Support EP48/96 for internode kernels and update some config. (#421)
parent
abba6add
Changes
3
Hide whitespace changes
Inline
Side-by-side
Showing
3 changed files
with
12 additions
and
6 deletions
+12
-6
csrc/kernels/launch.cuh
csrc/kernels/launch.cuh
+2
-0
deep_ep/buffer.py
deep_ep/buffer.py
+9
-5
tests/test_internode.py
tests/test_internode.py
+1
-1
No files found.
csrc/kernels/launch.cuh
View file @
c939644c
...
@@ -63,7 +63,9 @@ cfg.dynamicSmemBytes = smem_size;
...
@@ -63,7 +63,9 @@ cfg.dynamicSmemBytes = smem_size;
switch (num_ranks / NUM_MAX_NVL_PEERS) { \
switch (num_ranks / NUM_MAX_NVL_PEERS) { \
case 2: case_macro(2); \
case 2: case_macro(2); \
case 4: case_macro(4); \
case 4: case_macro(4); \
case 6: case_macro(6); \
case 8: case_macro(8); \
case 8: case_macro(8); \
case 12: case_macro(12); \
case 16: case_macro(16); \
case 16: case_macro(16); \
case 18: case_macro(18); \
case 18: case_macro(18); \
case 20: case_macro(20); \
case 20: case_macro(20); \
...
...
deep_ep/buffer.py
View file @
c939644c
...
@@ -235,9 +235,11 @@ class Buffer:
...
@@ -235,9 +235,11 @@ class Buffer:
8
:
Config
(
Buffer
.
num_sms
,
6
,
256
,
6
,
128
),
8
:
Config
(
Buffer
.
num_sms
,
6
,
256
,
6
,
128
),
16
:
Config
(
Buffer
.
num_sms
,
36
,
288
,
20
,
128
),
16
:
Config
(
Buffer
.
num_sms
,
36
,
288
,
20
,
128
),
24
:
Config
(
Buffer
.
num_sms
,
8
,
288
,
32
,
128
),
24
:
Config
(
Buffer
.
num_sms
,
8
,
288
,
32
,
128
),
32
:
Config
(
Buffer
.
num_sms
,
32
,
288
,
32
,
128
),
32
:
Config
(
Buffer
.
num_sms
,
32
,
288
,
8
,
128
),
64
:
Config
(
Buffer
.
num_sms
,
20
,
288
,
28
,
128
),
48
:
Config
(
Buffer
.
num_sms
,
32
,
288
,
8
,
128
),
128
:
Config
(
Buffer
.
num_sms
,
20
,
560
,
32
,
128
),
64
:
Config
(
Buffer
.
num_sms
,
32
,
288
,
8
,
128
),
96
:
Config
(
Buffer
.
num_sms
,
20
,
480
,
12
,
128
),
128
:
Config
(
Buffer
.
num_sms
,
20
,
560
,
12
,
128
),
144
:
Config
(
Buffer
.
num_sms
,
32
,
720
,
12
,
128
),
144
:
Config
(
Buffer
.
num_sms
,
32
,
720
,
12
,
128
),
160
:
Config
(
Buffer
.
num_sms
,
28
,
720
,
12
,
128
),
160
:
Config
(
Buffer
.
num_sms
,
28
,
720
,
12
,
128
),
}
}
...
@@ -264,8 +266,10 @@ class Buffer:
...
@@ -264,8 +266,10 @@ class Buffer:
16
:
Config
(
Buffer
.
num_sms
,
4
,
288
,
12
,
128
),
16
:
Config
(
Buffer
.
num_sms
,
4
,
288
,
12
,
128
),
24
:
Config
(
Buffer
.
num_sms
,
1
,
288
,
8
,
128
),
24
:
Config
(
Buffer
.
num_sms
,
1
,
288
,
8
,
128
),
32
:
Config
(
Buffer
.
num_sms
,
1
,
288
,
8
,
128
),
32
:
Config
(
Buffer
.
num_sms
,
1
,
288
,
8
,
128
),
64
:
Config
(
Buffer
.
num_sms
,
1
,
288
,
20
,
128
),
48
:
Config
(
Buffer
.
num_sms
,
1
,
288
,
8
,
128
),
128
:
Config
(
Buffer
.
num_sms
,
1
,
560
,
12
,
128
),
64
:
Config
(
Buffer
.
num_sms
,
1
,
288
,
8
,
128
),
96
:
Config
(
Buffer
.
num_sms
,
1
,
480
,
8
,
128
),
128
:
Config
(
Buffer
.
num_sms
,
1
,
560
,
8
,
128
),
144
:
Config
(
Buffer
.
num_sms
,
2
,
720
,
8
,
128
),
144
:
Config
(
Buffer
.
num_sms
,
2
,
720
,
8
,
128
),
160
:
Config
(
Buffer
.
num_sms
,
2
,
720
,
8
,
128
),
160
:
Config
(
Buffer
.
num_sms
,
2
,
720
,
8
,
128
),
}
}
...
...
tests/test_internode.py
View file @
c939644c
...
@@ -90,7 +90,7 @@ def test_main(args: argparse.Namespace, num_sms: int,
...
@@ -90,7 +90,7 @@ def test_main(args: argparse.Namespace, num_sms: int,
time
.
sleep
(
1
)
time
.
sleep
(
1
)
# Config
# Config
rdma_buffer_size
,
nvl_buffer_size
=
128
,
(
720
if
num_ranks
in
(
144
,
160
)
else
512
)
rdma_buffer_size
,
nvl_buffer_size
=
128
,
(
720
if
num_ranks
in
(
48
,
96
,
144
,
160
)
else
512
)
config
=
deep_ep
.
Config
(
num_sms
,
8
,
nvl_buffer_size
,
16
,
rdma_buffer_size
)
config
=
deep_ep
.
Config
(
num_sms
,
8
,
nvl_buffer_size
,
16
,
rdma_buffer_size
)
# Test dispatch
# Test dispatch
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment