Skip to content

[0-size Tensor No.355] Add 0-size Tensor support for unique_consecutive #73444

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

Closed
wants to merge 4 commits into from
Closed

[0-size Tensor No.355] Add 0-size Tensor support for unique_consecutive #73444

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
57f07d2
Add 0-size Tensor support for unique_consecutive
ccsuzzh Jun 18, 2025
9ddad45
fix codestyle
ccsuzzh Jun 18, 2025
7376b5b
fix bug
ccsuzzh Jun 19, 2025
73bb142
fix bug
ccsuzzh Jun 19, 2025
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
51 changes: 43 additions & 8 deletions paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/unary_infer_sym.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4351,14 +4351,49 @@ bool UniqueConsecutiveOpInferSymbolicShape(
return inverse_dims;
}();

infer_context->SetShapeOrDataForValue(
op->result(0), symbol::TensorShapeOrDataDimExprs{out_dims});
infer_context->SetShapeOrDataForValue(
op->result(1),
return_inverse ? symbol::TensorShapeOrDataDimExprs{inverse_dims} : empty);
infer_context->SetShapeOrDataForValue(
op->result(2),
return_counts ? symbol::TensorShapeOrDataDimExprs{counts_dims} : empty);
const auto &IsZero = [&](const symbol::DimExpr &dim_expr) {
if (dim_expr.isa<int64_t>()) {
return dim_expr.dyn_cast<int64_t>() == static_cast<int64_t>(0);
}
return false;
};
bool size_0 = false;
for (size_t i = 0; i < x_dims_sym.size(); i++) {
if (IsZero(x_dims_sym.at(i))) {
size_0 = true;
break;
}
}

if (size_0) {
if (axes.empty()) {
infer_context->SetShapeOrDataForValue(
op->result(0), symbol::TensorShapeOrDataDimExprs{});
infer_context->SetShapeOrDataForValue(
op->result(1),
return_inverse ? symbol::TensorShapeOrDataDimExprs{x_dims_sym}
: empty);
} else {
infer_context->SetShapeOrDataForValue(
op->result(0), symbol::TensorShapeOrDataDimExprs{x_dims_sym});
infer_context->SetShapeOrDataForValue(
op->result(1),
return_inverse ? symbol::TensorShapeOrDataDimExprs{} : empty);
}
infer_context->SetShapeOrDataForValue(
op->result(2),
return_counts ? symbol::TensorShapeOrDataDimExprs{} : empty);
} else {
infer_context->SetShapeOrDataForValue(
op->result(0), symbol::TensorShapeOrDataDimExprs{out_dims});
infer_context->SetShapeOrDataForValue(
op->result(1),
return_inverse ? symbol::TensorShapeOrDataDimExprs{inverse_dims}
: empty);
infer_context->SetShapeOrDataForValue(
op->result(2),
return_counts ? symbol::TensorShapeOrDataDimExprs{counts_dims} : empty);
}

return true;
}
Expand Down
16 changes: 16 additions & 0 deletions paddle/phi/infermeta/unary.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5918,6 +5918,22 @@ void UniqueConsecutiveInferMeta(const MetaTensor& x,
if (return_counts) {
counts->set_dims({-1});
}
if (x.numel() == 0) {
if (axis.empty()) {
out->set_dims({0});
if (return_inverse) {
index->set_dims(in_dims);
}
} else {
out->set_dims(in_dims);
if (return_inverse) {
index->set_dims({0});
}
}
if (return_counts) {
counts->set_dims({0});
Comment on lines +5922 to +5934
Copy link
Contributor

@DanielSun11 DanielSun11 Jun 20, 2025

Choose a reason for hiding this comment

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

这个地方还是有些疑问。
out->set_dims({0}) 设置完后，tensor的shape是不是[0] ? shape [0] 和shape[ ] 是两个不同的shape，shape [0]的dims().size() 为1，是1d Tensor，尽管是0-size。shape [ ]的dims().size() 为0，是0d Tensor也是个标量。 试试out->set_dims({}),让tensor的shape为[ ]
请把

import torch
x = torch.randn([0,1,2])
output, inverse, counts = torch.unique_consecutive(x, return_inverse=True,return_counts=True)
print("x ",x)
print("output ",output)
print("inverse ",inverse)
print("counts ",counts)

转化为paddle的代码，跑一下，将结果贴到PR描述里。

}
}
}

void UniqueInferMeta(const MetaTensor& x,
Expand Down
11 changes: 11 additions & 0 deletions paddle/phi/kernels/cpu/unique_consecutive_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,17 @@ void UniqueConsecutiveKernel(const Context& dev_ctx,
DenseTensor* out,
DenseTensor* index,
DenseTensor* counts) {
if (out && out->numel() == 0) {
dev_ctx.template Alloc<T>(out);
if (return_inverse) {
dev_ctx.Alloc(index, dtype);
}
if (return_counts) {
dev_ctx.Alloc(counts, dtype);
}

return;
}
if (dtype == phi::DataType::INT32) {
PADDLE_ENFORCE_LE(
x.numel(),
Expand Down
11 changes: 11 additions & 0 deletions paddle/phi/kernels/gpu/unique_consecutive_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,17 @@ void UniqueConsecutiveKernel(const Context& dev_ctx,
DenseTensor* out,
DenseTensor* index,
DenseTensor* counts) {
if (out && out->numel() == 0) {
dev_ctx.template Alloc<T>(out);
if (return_inverse) {
dev_ctx.Alloc(index, dtype);
}
if (return_counts) {
dev_ctx.Alloc(counts, dtype);
}

return;
}
if (dtype == phi::DataType::INT32) {
PADDLE_ENFORCE_LE(
x.numel() + 1,
Expand Down
33 changes: 33 additions & 0 deletions test/legacy_test/test_unique_consecutive_op.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -372,6 +372,39 @@ def test_check_output(self):
self.check_output(check_pir=True, check_symbol_infer=False)


class TestUniqueConsecutiveZeroSize(TestUniqueConsecutiveOp):
"""ZeroSize input"""

def config(self):
self.x_size = (0, 2, 4)
self.x_range = 20
self.return_inverse = True
self.return_counts = True
self.python_api = paddle.unique_consecutive

def setUp(self):
self.init_kernel_type()
self.config()
self.op_type = "unique_consecutive"
x = np.random.randint(self.x_range, size=self.x_size).astype(self.dtype)
result, inverse, counts = reference_unique_consecutive(
x, self.return_inverse, self.return_counts
)
result = np.array(result).astype(self.dtype)
inverse = inverse.astype(self.dtype).reshape(x.shape)
counts = counts.astype(self.dtype)
self.inputs = {
'X': x,
}
self.attrs = {
'return_inverse': self.return_inverse,
'return_counts': self.return_counts,
'dtype': paddle.int32,
}
self.python_out_sig = ["Out", "Index", "Counts"]
self.outputs = {'Out': result, 'Index': inverse, 'Counts': counts}


if __name__ == "__main__":
paddle.enable_static()
unittest.main()
Loading