Skip to content

【Hackathon 9th No.17、18】 Fix accuracy diff for conv2d_transpose API with NHWC format #75141

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
Sep 10, 2025
Merged

【Hackathon 9th No.17、18】 Fix accuracy diff for conv2d_transpose API with NHWC format #75141

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
56 changes: 36 additions & 20 deletions paddle/phi/kernels/funcs/im2col_cfo_cpu.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -210,11 +210,15 @@ inline void im2col_sh1sw1dh1dw1ph1pw1(const phi::DenseTensor& im,
std::memcpy(dst_data + plw, src_data, copy_size);
} else {
for (int kow = 0; kow < output_width - plw - prw; ++kow) {
dst_data[plw + kow] =
im_data[(((oh - plh > 0 ? oh - plh : 0) + kh) * im_width +
kow) *
im_channels +
ic];
int im_row = oh - plh + kh;
int im_col = kow;
if (im_row >= 0 && im_row < im_height && im_col >= 0 &&
im_col < im_width) {
dst_data[plw + kow] =
im_data[(im_row * im_width + im_col) * im_channels + ic];
} else {
dst_data[plw + kow] = static_cast<T>(0);
}
}
}
dst_data = dst_data + col_matrix_width;
Expand Down Expand Up @@ -269,11 +273,15 @@ inline void im2col_sh1sw1dh1dw1ph1pw1(const phi::DenseTensor& im,
sizeof(T) * (output_width - (plw - kw)));
} else {
for (int kow = 0; kow < output_width - (plw - kw); ++kow) {
dst_data[plw - kw + kow] =
im_data[(((oh - plh > 0 ? oh - plh : 0) + kh) * im_width +
kow) *
im_channels +
ic];
int im_row = oh - plh + kh;
int im_col = kow;
if (im_row >= 0 && im_row < im_height && im_col >= 0 &&
im_col < im_width) {
dst_data[plw - kw + kow] =
im_data[(im_row * im_width + im_col) * im_channels + ic];
} else {
dst_data[plw - kw + kow] = static_cast<T>(0);
}
}
}
dst_data = dst_data + col_matrix_width;
Expand All @@ -284,11 +292,15 @@ inline void im2col_sh1sw1dh1dw1ph1pw1(const phi::DenseTensor& im,
dst_data, src_data + (kw - plw), sizeof(T) * output_width);
} else {
for (int kow = 0; kow < output_width; ++kow) {
dst_data[kow] =
im_data[(((oh - plh > 0 ? oh - plh : 0) + kh) * im_width +
kw - plw + kow) *
im_channels +
ic];
int im_row = oh - plh + kh;
int im_col = kw - plw + kow;
if (im_row >= 0 && im_row < im_height && im_col >= 0 &&
im_col < im_width) {
dst_data[kow] =
im_data[(im_row * im_width + im_col) * im_channels + ic];
} else {
dst_data[kow] = static_cast<T>(0);
}
}
}
dst_data = dst_data + col_matrix_width;
Expand All @@ -301,11 +313,15 @@ inline void im2col_sh1sw1dh1dw1ph1pw1(const phi::DenseTensor& im,
sizeof(T) * (output_width - i));
} else {
for (int kow = 0; kow < output_width - i; ++kow) {
dst_data[kow] =
im_data[(((oh - plh > 0 ? oh - plh : 0) + kh) * im_width +
kw - plw + kow) *
im_channels +
ic];
int im_row = oh - plh + kh;
int im_col = kw - plw + kow;
if (im_row >= 0 && im_row < im_height && im_col >= 0 &&
im_col < im_width) {
dst_data[kow] =
im_data[(im_row * im_width + im_col) * im_channels + ic];
} else {
dst_data[kow] = static_cast<T>(0);
}
}
}
dst_data = dst_data + col_matrix_width;
Expand Down
24 changes: 24 additions & 0 deletions test/legacy_test/test_conv2d_transpose_op.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1575,5 +1575,29 @@ def init_data(self):
self.np_out = np.zeros([4, 0, 6, 6])


class TestWithSAMEPad_NHWC(TestConv2DTransposeOp):
def init_test_case(self):
self.stride = [1, 1]
self.dilations = [1, 1]
self.groups = 1
self.input_size = [1, 3, 3, 1] # NHWC
f_c = self.input_size[-1]
self.filter_size = [f_c, 2, 3, 3]
self.data_format = 'NHWC'
self.padding_algorithm = 'SAME'


class TestWithSAMEPadGroups_NHWC(TestConv2DTransposeOp):
def init_test_case(self):
self.stride = [1, 1]
self.dilations = [1, 1]
self.groups = 2
self.input_size = [1, 3, 3, 2] # NHWC
f_c = self.input_size[-1]
self.filter_size = [f_c, 1, 3, 3]
self.data_format = 'NHWC'
self.padding_algorithm = 'SAME'


if __name__ == '__main__':
unittest.main()