sparse element-wise multiplication (#323)

* sparse element-wise multiplication

* update

* update

* update

* update

---------

Co-authored-by: Andreas Bergmeister <[email protected]>
Co-authored-by: rusty1s <[email protected]>

Author: 3 people

.github/workflows/testing.yml

@@ -48,12 +48,13 @@ jobs:
 
       - name: Install main package
         run: |
-          pip install -e .[test]
+          python setup.py develop
         env:
           WITH_METIS: 1
 
       - name: Run test-suite
         run: |
+          pip install pytest pytest-cov
           pytest --cov --cov-report=xml
 
       - name: Upload coverage

test/test_mul.py

@@ -0,0 +1,53 @@
+from itertools import product
+
+import pytest
+import torch
+
+from torch_sparse import SparseTensor, mul
+from torch_sparse.testing import devices, dtypes, tensor
+
+
+@pytest.mark.parametrize('dtype,device', product(dtypes, devices))
+def test_sparse_sparse_mul(dtype, device):
+    rowA = torch.tensor([0, 0, 1, 2, 2], device=device)
+    colA = torch.tensor([0, 2, 1, 0, 1], device=device)
+    valueA = tensor([1, 2, 4, 1, 3], dtype, device)
+    A = SparseTensor(row=rowA, col=colA, value=valueA)
+
+    rowB = torch.tensor([0, 0, 1, 2, 2], device=device)
+    colB = torch.tensor([1, 2, 2, 1, 2], device=device)
+    valueB = tensor([2, 3, 1, 2, 4], dtype, device)
+    B = SparseTensor(row=rowB, col=colB, value=valueB)
+
+    C = A * B
+    rowC, colC, valueC = C.coo()
+
+    assert rowC.tolist() == [0, 2]
+    assert colC.tolist() == [2, 1]
+    assert valueC.tolist() == [6, 6]
+
+    @torch.jit.script
+    def jit_mul(A: SparseTensor, B: SparseTensor) -> SparseTensor:
+        return mul(A, B)
+
+    jit_mul(A, B)
+
+
+@pytest.mark.parametrize('dtype,device', product(dtypes, devices))
+def test_sparse_sparse_mul_empty(dtype, device):
+    rowA = torch.tensor([0], device=device)
+    colA = torch.tensor([1], device=device)
+    valueA = tensor([1], dtype, device)
+    A = SparseTensor(row=rowA, col=colA, value=valueA)
+
+    rowB = torch.tensor([1], device=device)
+    colB = torch.tensor([0], device=device)
+    valueB = tensor([2], dtype, device)
+    B = SparseTensor(row=rowB, col=colB, value=valueB)
+
+    C = A * B
+    rowC, colC, valueC = C.coo()
+
+    assert rowC.tolist() == []
+    assert colC.tolist() == []
+    assert valueC.tolist() == []

torch_sparse/mul.py

@@ -1,27 +1,83 @@
 from typing import Optional
 
 import torch
+from torch import Tensor
 from torch_scatter import gather_csr
+
 from torch_sparse.tensor import SparseTensor
 
 
-def mul(src: SparseTensor, other: torch.Tensor) -> SparseTensor:
-    rowptr, col, value = src.csr()
-    if other.size(0) == src.size(0) and other.size(1) == 1:  # Row-wise...
-        other = gather_csr(other.squeeze(1), rowptr)
-        pass
-    elif other.size(0) == 1 and other.size(1) == src.size(1):  # Col-wise...
-        other = other.squeeze(0)[col]
-    else:
-        raise ValueError(
-            f'Size mismatch: Expected size ({src.size(0)}, 1, ...) or '
-            f'(1, {src.size(1)}, ...), but got size {other.size()}.')
+@torch.jit._overload  # noqa: F811
+def mul(src, other):  # noqa: F811
+    # type: (SparseTensor, Tensor) -> SparseTensor
+    pass
 
-    if value is not None:
-        value = other.to(value.dtype).mul_(value)
+
+@torch.jit._overload  # noqa: F811
+def mul(src, other):  # noqa: F811
+    # type: (SparseTensor, SparseTensor) -> SparseTensor
+    pass
+
+
+def mul(src, other):  # noqa: F811
+    if isinstance(other, Tensor):
+        rowptr, col, value = src.csr()
+        if other.size(0) == src.size(0) and other.size(1) == 1:  # Row-wise...
+            other = gather_csr(other.squeeze(1), rowptr)
+            pass
+        # Col-wise...
+        elif other.size(0) == 1 and other.size(1) == src.size(1):
+            other = other.squeeze(0)[col]
+        else:
+            raise ValueError(
+                f'Size mismatch: Expected size ({src.size(0)}, 1, ...) or '
+                f'(1, {src.size(1)}, ...), but got size {other.size()}.')
+
+        if value is not None:
+            value = other.to(value.dtype).mul_(value)
+        else:
+            value = other
+        return src.set_value(value, layout='coo')
+
+    assert isinstance(other, SparseTensor)
+
+    if not src.is_coalesced():
+        raise ValueError("The `src` tensor is not coalesced")
+    if not other.is_coalesced():
+        raise ValueError("The `other` tensor is not coalesced")
+
+    rowA, colA, valueA = src.coo()
+    rowB, colB, valueB = other.coo()
+
+    row = torch.cat([rowA, rowB], dim=0)
+    col = torch.cat([colA, colB], dim=0)
+
+    if valueA is not None and valueB is not None:
+        value = torch.cat([valueA, valueB], dim=0)
     else:
-        value = other
-    return src.set_value(value, layout='coo')
+        raise ValueError('Both sparse tensors must contain values')
+
+    M = max(src.size(0), other.size(0))
+    N = max(src.size(1), other.size(1))
+    sparse_sizes = (M, N)
+
+    # Sort indices:
+    idx = col.new_full((col.numel() + 1, ), -1)
+    idx[1:] = row * sparse_sizes[1] + col
+    perm = idx[1:].argsort()
+    idx[1:] = idx[1:][perm]
+
+    row, col, value = row[perm], col[perm], value[perm]
+
+    valid_mask = idx[1:] == idx[:-1]
+    valid_idx = valid_mask.nonzero().view(-1)
+
+    return SparseTensor(
+        row=row[valid_mask],
+        col=col[valid_mask],
+        value=value[valid_idx - 1] * value[valid_idx],
+        sparse_sizes=sparse_sizes,
+    )
 
 
 def mul_(src: SparseTensor, other: torch.Tensor) -> SparseTensor:
@@ -43,8 +99,11 @@ def mul_(src: SparseTensor, other: torch.Tensor) -> SparseTensor:
     return src.set_value_(value, layout='coo')
 
 
-def mul_nnz(src: SparseTensor, other: torch.Tensor,
-            layout: Optional[str] = None) -> SparseTensor:
+def mul_nnz(
+    src: SparseTensor,
+    other: torch.Tensor,
+    layout: Optional[str] = None,
+) -> SparseTensor:
     value = src.storage.value()
     if value is not None:
         value = value.mul(other.to(value.dtype))
@@ -53,8 +112,11 @@ def mul_nnz(src: SparseTensor, other: torch.Tensor,
     return src.set_value(value, layout=layout)
 
 
-def mul_nnz_(src: SparseTensor, other: torch.Tensor,
-             layout: Optional[str] = None) -> SparseTensor:
+def mul_nnz_(
+    src: SparseTensor,
+    other: torch.Tensor,
+    layout: Optional[str] = None,
+) -> SparseTensor:
     value = src.storage.value()
     if value is not None:
         value = value.mul_(other.to(value.dtype))