8000 add dynamic shape support for sin/cos/cat by lanluo-nvidia · Pull Request #2887 · pytorch/TensorRT · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

add dynamic shape support for sin/cos/cat #2887

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 4 commits into from
Jun 10, 2024
Merged

add dynamic shape support for sin/cos/cat #2887

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
4e5334a
add dynamic shape support for sin/cos
lanluo-nvidia Jun 4, 2024
9b05272
add dynamic shape support for cat/sin/cos
lanluo-nvidia Jun 4, 2024
cacb23b
Merge branch 'lluo/dynamic_shapes_sin_cos' of github.com:pytorch/Tens…
lanluo-nvidia Jun 5, 2024
ed18394
Replace shape_ranges to min_shape/opt_shape/max_shape
lanluo-nvidia Jun 5, 2024
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
6 changes: 3 additions & 3 deletions py/torch_tensorrt/dynamo/conversion/aten_ops_converters.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -219,7 +219,7 @@ def aten_ops_group_norm(
)


@dynamo_tensorrt_converter(torch.ops.aten.cat.default)
@dynamo_tensorrt_converter(torch.ops.aten.cat.default, supports_dynamic_shapes=True)
def aten_ops_cat(
ctx: ConversionContext,
target: Target,
Expand Down Expand Up @@ -1327,7 +1327,7 @@ def aten_ops_abs(
)


@dynamo_tensorrt_converter(torch.ops.aten.sin.default)
@dynamo_tensorrt_converter(torch.ops.aten.sin.default, supports_dynamic_shapes=True)
def aten_ops_sin(
ctx: ConversionContext,
target: Target,
Expand All @@ -1344,7 +1344,7 @@ def aten_ops_sin(
)


@dynamo_tensorrt_converter(torch.ops.aten.cos.default)
@dynamo_tensorrt_converter(torch.ops.aten.cos.default, supports_dynamic_shapes=True)
def aten_ops_cos(
ctx: ConversionContext,
target: Target,
Expand Down
50 changes: 42 additions & 8 deletions tests/py/dynamo/conversion/test_cat_aten.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,14 +38,16 @@ def forward(self, x, y):

input_specs = [
Input(
shape=(16, -1, 3),
dtype=torch.float32,
shape_ranges=[((16, 2, 3), (16, 3, 3), (16, 32, 3))],
min_shape=(16, 2, 3),
opt_shape=(16, 3, 3),
max_shape=(16, 32, 3),
),
Input(
shape=(16, -1, 3),
dtype=torch.float32,
shape_ranges=[((16, 2, 3), (16, 16, 3), (16, 32, 3))],
min_shape=(16, 2, 3),
opt_shape=(16, 16, 3),
max_shape=(16, 32, 3),
),
]
self.run_test_with_dynamic_shape(
Expand All @@ -71,14 +73,46 @@ def forward(self, x, y):

input_specs = [
Input(
shape=(-1, 16, 3),
dtype=torch.float32,
shape_ranges=[((2, 16, 3), (3, 16, 3), (32, 16, 3))],
min_shape=(2, 16, 3),
opt_shape=(3, 16, 3),
max_shape=(32, 16, 3),
),
Input(
shape=(-1, 16, 3),
dtype=torch.float32,
shape_ranges=[((2, 16, 3), (3, 16, 3), (32, 16, 3))],
min_shape=(2, 16, 3),
opt_shape=(3, 16, 3),
max_shape=(32, 16, 3),
),
]
self.run_test_with_dynamic_shape(
Cat(),
input_specs,
)

@parameterized.expand(
[
("pos", 1),
("neg", -2),
]
)
def test_cat_dynamic_shape_dim(self, _, dim):
class Cat(nn.Module):
def forward(self, x, y):
return torch.ops.aten.cat.default((x, y), dim)

input_specs = [
Input(
dtype=torch.float32,
min_shape=(2, 1, 1),
opt_shape=(3, 1, 2),
max_shape=(4, 1, 3),
),
Input(
dtype=torch.float32,
min_shape=(2, 2, 1),
opt_shape=(3, 3, 2),
max_shape=(4, 4, 3),
),
]
self.run_test_with_dynamic_shape(
Expand Down
49 changes: 49 additions & 0 deletions tests/py/dynamo/conversion/test_cos_aten.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@
import torch.nn as nn
from parameterized import parameterized
from torch.testing._internal.common_utils import run_tests
from torch_tensorrt import Input

from .harness import DispatchTestCase

Expand Down Expand Up @@ -44,6 +45,54 @@ def forward(self, input):
inputs,
)

@parameterized.expand(
[
(
"3d_dim_dtype_int32",
(3, 2, 1),
(3, 2, 3),
(3, 3, 4),
torch.int32,
torch.float32,
),
(
"2d_dim_dtype_float16",
(1, 1),
(2, 2),
(4, 4),
torch.float16,
torch.float16,
),
(
"3d_dim_dtype_float",
(1, 1, 1),
(1, 2, 3),
(3, 3, 3),
torch.float,
torch.float,
),
]
)
def test_dynamic_shape_cos(
self, _, min_shape, opt_shape, max_shape, type, output_type
):
class cos(nn.Module):
def forward(self, input):
return torch.ops.aten.cos.default(input)

input_specs = [
Input(
min_shape=min_shape,
opt_shape=opt_shape,
max_shape=max_shape,
dtype=type,
),
]

self.run_test_with_dynamic_shape(
cos(), input_specs, output_dtypes=[output_type]
)


if __name__ == "__main__":
run_tests()
49 changes: 49 additions & 0 deletions tests/py/dynamo/conversion/test_sin_aten.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@
import torch.nn as nn
from parameterized import parameterized
from torch.testing._internal.common_utils import run_tests
from torch_tensorrt import Input

from .harness import DispatchTestCase

Expand Down Expand Up @@ -44,6 +45,54 @@ def forward(self, input):
inputs,
)

@parameterized.expand(
[
(
"3d_dim_dtype_int32",
(3, 2, 1),
(3, 2, 3),
(3, 3, 4),
torch.int32,
torch.float32,
),
(
"2d_dim_dtype_float16",
(1, 1),
(2, 2),
(4, 4),
torch.float16,
torch.float16,
),
(
"3d_dim_dtype_float",
(1, 1, 1),
(1, 2, 3),
(3, 3, 3),
torch.float,
torch.float,
),
]
)
def test_dynamic_shape_sin(
self, _, min_shape, opt_shape, max_shape, type, output_type
):
class sin(nn.Module):
def forward(self, input):
return torch.ops.aten.sin.default(input)

input_specs = [
Input(
min_shape=min_shape,
opt_shape=opt_shape,
max_shape=max_shape,
dtype=type,
),
]

self.run_test_with_dynamic_shape(
sin(), input_specs, output_dtypes=[output_type]
)


if __name__ == "__main__":
run_tests()
Loading
0