Skip to content

[Frontend][PaddlePaddle] Support conv2d when data_format is NHWC #16616

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
jiangjiajun merged 3 commits into from
Feb 22, 2024
Merged

[Frontend][PaddlePaddle] Support conv2d when data_format is NHWC #16616

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c2bf9fb
support conv2d when data_format is NHWC
Zheng-Bicheng Feb 21, 2024
f6223d1
modify the annotation
Zheng-Bicheng Feb 21, 2024
0f89323
Merge branch 'apache:main' into support_nhwc
Zheng-Bicheng Feb 22, 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
18 changes: 18 additions & 0 deletions python/tvm/relay/frontend/paddlepaddle.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -314,6 +314,7 @@ def convert_conv2d(g, op, block):
strides = op.attr("strides")

kernel = g.get_node(op.input("Filter")[0])
kernel_layout = "OIHW"
input_x = g.get_node(op.input("Input")[0])
data_layout = op.attr("data_format")
out_channels, _, k_h, k_w = infer_shape(kernel)
Expand All @@ -335,6 +336,16 @@ def convert_conv2d(g, op, block):
msg = f'Value {padding_algorithm} in attribute "padding" of operator Conv is not "valid."'
raise tvm.error.OpAttributeInvalid(msg)

if data_layout == "NHWC":
kernel_layout = "HWIO"
# PaddlePaddle wieght layout is "OIHW", tvm need "HWIO" when op data_format is "NHWC".
kernel_data = g.get_params(op.input("Filter")[0])
kernel_data = kernel_data.asnumpy()
kernel_data = kernel_data.transpose((2, 3, 1, 0))
kernel_data = _nd.array(kernel_data)
g.modify_node(op.input("Filter")[0], kernel_data)
kernel = g.get_node(op.input("Filter")[0])

out = _op.nn.conv2d(
input_x,
kernel,
Expand All @@ -345,6 +356,7 @@ def convert_conv2d(g, op, block):
channels=out_channels,
kernel_size=[k_h, k_w],
data_layout=data_layout,
kernel_layout=kernel_layout,
)
g.add_node(op.output("Output")[0], out)

Expand Down Expand Up @@ -2915,6 +2927,12 @@ def add_node(self, name, node):

self.nodes[name] = fold_constant(node)

def modify_node(self, name, params):
"""modify node from graph"""

self.params[name] = params
self.nodes[name] = new_var(name, shape=params.shape, dtype=params.dtype)

def get_params(self, name=None):
"""Get params from graph."""

Expand Down