Skip to content
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

Jump to bottom

Fix roll kernel gpu bug. #52012

Merged
merged 4 commits into from
Mar 24, 2023
Merged

Fix roll kernel gpu bug. #52012

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
f6196e5
add corresponding ut
FeixLiu Mar 22, 2023
774ecc1
update shape
FeixLiu Mar 22, 2023
06a7aba
fix the bug
FeixLiu Mar 23, 2023
10f5de8
update the ut
FeixLiu Mar 23, 2023
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
2 changes: 1 addition & 1 deletion paddle/phi/kernels/gpu/roll_kernel_impl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,7 @@ __global__ void RollCudaKernel(const T* input,

#pragma unroll
for (size_t i = 0; i < Rank; i++) {
new_dim_idx = (idx / strides[i]) % sizes[i] + shifts[i];
new_dim_idx = (output_idx / strides[i]) % sizes[i] + shifts[i];
if (new_dim_idx >= sizes[i]) {
output_idx += (shifts[i] - sizes[i]) * strides[i];
} else {
Expand Down
36 changes: 36 additions & 0 deletions python/paddle/fluid/tests/unittests/test_roll_op.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,14 @@ def init_dtype_type(self):
self.axis = [-1, -2]


class TestRollOpCase3(TestRollOp):
def init_dtype_type(self):
self.dtype = np.float32
self.x_shape = (11, 11)
self.shifts = [1, 1]
self.axis = [-1, 1]


class TestRollFP16OP(TestRollOp):
def init_dtype_type(self):
self.dtype = np.float16
Expand All @@ -77,6 +85,14 @@ def init_dtype_type(self):
self.axis = [-1, -2]


class TestRollFP16OpCase3(TestRollOp):
def init_dtype_type(self):
self.dtype = np.float16
self.x_shape = (11, 11)
self.shifts = [1, 1]
self.axis = [-1, 1]


@unittest.skipIf(
not core.is_compiled_with_cuda()
or not core.is_bfloat16_supported(core.CUDAPlace(0)),
Expand Down Expand Up @@ -117,6 +133,26 @@ def test_check_grad_normal(self):
self.check_grad_with_place(self.place, ['X'], 'Out', check_eager=True)


@unittest.skipIf(
not core.is_compiled_with_cuda()
or not core.is_bfloat16_supported(core.CUDAPlace(0)),
"core is not complied with CUDA and not support the bfloat16",
)
class TestRollBF16OpCase3(TestRollOp):
def init_dtype_type(self):
self.dtype = np.uint16
self.x_shape = (11, 11)
self.shifts = [1, 1]
self.axis = [-1, 1]
self.place = core.CUDAPlace(0)

def test_check_output(self):
self.check_output_with_place(self.place, check_eager=True)

def test_check_grad_normal(self):
self.check_grad_with_place(self.place, ['X'], 'Out', check_eager=True)


class TestRollAPI(unittest.TestCase):
def input_data(self):
self.data_x = np.array(
Expand Down