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.

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

[Frontend][Pytorch]Add Pytorch advanced indexing #6318

Merged
merged 3 commits into from
Aug 22, 2020
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
53 changes: 50 additions & 3 deletions python/tvm/relay/frontend/pytorch.py
Original file line number Diff line number Diff line change
Expand Up @@ -274,16 +274,18 @@ def _impl(inputs, input_types):
end[dim] = min(end[dim], int(inputs[3]))
else:
if isinstance(inputs[3], _expr.Call):
end[dim] = np.asscalar(_infer_value(inputs[3], {}).asnumpy().astype(np.int))
target_end = np.asscalar(_infer_value(inputs[3], {}).asnumpy().astype(np.int))
else:
end[dim] = inputs[3]
target_end = inputs[3]

end[dim] = min(end[dim], target_end)

strides.append(int(inputs[4]))
return _op.transform.strided_slice(data,
begin=_expr.const(begin),
end=_expr.const(end),
strides=_expr.const(strides),
slice_mode="size")
slice_mode="end")
return _impl

def _split():
Expand Down Expand Up @@ -1755,6 +1757,50 @@ def _impl(inputs, input_types):
return _impl


def _index():
def _impl(inputs, input_types):
data = inputs[0]
indices = []
raw_indices = []
max_indices_len = -1
for index in inputs[1]:
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

seems these two for loops are able to be merged

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

max_indices_len needs to be fetched first.

if not isinstance(index, _expr.Constant):
try:
index = _expr.const(_infer_value(index, {}))
except Exception:
raise RuntimeError("Only supports constant indices for "
"pytorch advanced indexing ")
raw_indices.append(index)
cindex_len = index.data.shape[0]
if cindex_len > max_indices_len:
max_indices_len = cindex_len

for index in raw_indices:
cnp = index.data.asnumpy()
cindex_len = cnp.shape[0]
if cindex_len < max_indices_len:
cnp = np.tile(cnp, max_indices_len // cindex_len)
indices.append(cnp)

ret = []
slice_map = {}
for i in range(indices[0].shape[0]):
tmp = data
current_indices = []
for index in indices:
current_indices.append(index[i])
index_key = tuple(current_indices)
if index_key in slice_map:
tmp = slice_map[index_key]
else:
tmp = _op.take(tmp, _expr.const(index[i]), axis=0)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

out of curiosity, would this introduce many take and tile ops?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In most cases it should be fine, but we can definitely improve it by adding a new topi implementation.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I am okay with this for now. We can implement a topi op if when perf is not good.

slice_map[index_key] = tmp
ret.append(_op.expand_dims(tmp, axis=0))

return _op.concatenate(ret, axis=0)
return _impl


def _meshgrid():
def _impl(inputs, input_types):
data = inputs[0]
Expand Down Expand Up @@ -2060,6 +2106,7 @@ def _get_convert_map(prelude):
"aten::type_as" : _type_as(),
"aten::gather" : _gather(),
"aten::index_select" : _select(),
"aten::index" : _index(),
}
return convert_map

Expand Down
24 changes: 22 additions & 2 deletions tests/python/frontend/pytorch/test_forward.py
Original file line number Diff line number Diff line change
Expand Up @@ -1202,13 +1202,13 @@ def forward(self, *args):

class Slice2(Module):
def forward(self, *args):
return args[0][0, :, :, :]
return args[0][0, :, :-3, :]

class Slice3(Module):
def forward(self, *args):
x0 = torch.tensor(2) - torch.tensor(1)
x1 = torch.tensor(3) + torch.tensor(1)
return args[0][:, x0:, :x1, :]
return args[0][:, x0:, 1:x1, :]

input_data = torch.rand(input_shape).float()
verify_model(Slice1().float().eval(), input_data=input_data)
Expand Down Expand Up @@ -2607,6 +2607,25 @@ def forward(self, *args):
verify_model(MatMul1().float().eval(), input_data=[tensor1, tensor2])


def test_forward_index():
torch.set_grad_enabled(False)
input_shape = [3, 4, 5, 6]

class Index0(Module):
def forward(self, x):
return x[[0, 1], [0, 2], :2, 4]

input_data = torch.rand(input_shape).float()
verify_model(Index0().eval(), input_data=input_data)

class Index1(Module):
def forward(self, x):
return x[[0], [1, 2, 3, 0], [3, 1, 2, 2], [4, 2, 1, 0]]

input_data = torch.rand(input_shape).float()
verify_model(Index1().eval(), input_data=input_data)


def test_forward_pretrained_bert_base_uncased():
######################################################################
# This is an example how to run BERT models using TVM
Expand Down Expand Up @@ -2846,6 +2865,7 @@ def test_forward_pretrained_bert_base_uncased():
test_adaptive_pool3d()
test_conv3d()
test_conv3d_transpose()
test_forward_index()

# Model tests
test_resnet18()
Expand Down