feat: add torch.Tensor frontend masked_scatter and masked_scatter_ (#…

…28783) Co-authored-by: Jin Wang <[email protected]> Co-authored-by: Sam-Armstrong <[email protected]>
ivy-llc · Jul 10, 2024 · 8af3ae7 · 8af3ae7
1 parent e4539a2
commit 8af3ae7
Show file tree

Hide file tree

Showing 2 changed files with 104 additions and 0 deletions.
diff --git a/ivy/functional/frontends/torch/tensor.py b/ivy/functional/frontends/torch/tensor.py
@@ -1077,6 +1077,23 @@ def masked_fill_(self, mask, value):
     def masked_select(self, mask):
         return torch_frontend.masked_select(self, mask)
 
+    def masked_scatter(self, mask, source):
+        flat_self = torch_frontend.flatten(self.clone())
+        flat_mask = torch_frontend.flatten(mask)
+        flat_source = torch_frontend.flatten(source)
+        indices = torch_frontend.squeeze(torch_frontend.nonzero(flat_mask), -1)
+        flat_self.scatter_(0, indices, flat_source[:indices.shape[0]])
+        return flat_self.reshape(self.shape)
+
+    def masked_scatter_(self, mask, source):
+        flat_self = torch_frontend.flatten(self.clone())
+        flat_mask = torch_frontend.flatten(mask)
+        flat_source = torch_frontend.flatten(source)
+        indices = torch_frontend.squeeze(torch_frontend.nonzero(flat_mask), -1)
+        flat_self.scatter_(0, indices, flat_source[:indices.shape[0]])
+        self.ivy_array = flat_self.reshape(self.shape).ivy_array
+        return self
+
     @with_unsupported_dtypes({"2.2 and below": ("float16", "bfloat16")}, "torch")
     def index_add_(self, dim, index, source, *, alpha=1):
         self.ivy_array = torch_frontend.index_add(

diff --git a/ivy_tests/test_ivy/test_frontends/test_torch/test_tensor.py b/ivy_tests/test_ivy/test_frontends/test_torch/test_tensor.py
@@ -332,6 +332,24 @@ def _masked_fill_helper(draw):
     return dtypes[0], xs[0], cond, fill_value
 
 
+@st.composite
+def _masked_scatter_helper(draw):
+    shape = draw(helpers.get_shape(min_num_dims=1, min_dim_size=1))
+    dtypes, xs = draw(
+        helpers.dtype_and_values(
+            available_dtypes=helpers.get_dtypes("valid"),
+            num_arrays=2,
+            shape=shape,
+            shared_dtype=True,
+            large_abs_safety_factor=16,
+            small_abs_safety_factor=16,
+            safety_factor_scale="log",
+        )
+    )
+    mask = draw(helpers.array_values(dtype="bool", shape=shape))
+    return dtypes[0], xs[0], mask, xs[1]
+
+
 @st.composite
 def _repeat_helper(draw):
     shape = draw(
@@ -9345,6 +9363,75 @@ def test_torch_masked_select(
         on_device=on_device,
     )
 
+# masked_scatter
+@handle_frontend_method(
+    class_tree=CLASS_TREE,
+    init_tree="torch.tensor",
+    method_name="masked_scatter",
+    dtype_x_mask_val=_masked_scatter_helper(),
+)
+def test_torch_masked_scatter(
+    dtype_x_mask_val,
+    frontend_method_data,
+    init_flags,
+    method_flags,
+    frontend,
+    on_device,
+    backend_fw,
+):
+    dtype, x, mask, val = dtype_x_mask_val
+    helpers.test_frontend_method(
+        init_input_dtypes=[dtype],
+        backend_to_test=backend_fw,
+        init_all_as_kwargs_np={
+            "data": x,
+        },
+        method_input_dtypes=["bool", dtype],
+        method_all_as_kwargs_np={
+            "mask": mask,
+            "source": val,
+        },
+        frontend_method_data=frontend_method_data,
+        init_flags=init_flags,
+        method_flags=method_flags,
+        frontend=frontend,
+        on_device=on_device,
+    )
+
+# masked_scatter_
+@handle_frontend_method(
+    class_tree=CLASS_TREE,
+    init_tree="torch.tensor",
+    method_name="masked_scatter_",
+    dtype_x_mask_val=_masked_scatter_helper(),
+)
+def test_torch_masked_scatter_(
+    dtype_x_mask_val,
+    frontend_method_data,
+    init_flags,
+    method_flags,
+    frontend,
+    on_device,
+    backend_fw,
+):
+    dtype, x, mask, val = dtype_x_mask_val
+    helpers.test_frontend_method(
+        init_input_dtypes=[dtype],
+        backend_to_test=backend_fw,
+        init_all_as_kwargs_np={
+            "data": x,
+        },
+        method_input_dtypes=["bool", dtype],
+        method_all_as_kwargs_np={
+            "mask": mask,
+            "source": val,
+        },
+        frontend_method_data=frontend_method_data,
+        init_flags=init_flags,
+        method_flags=method_flags,
+        frontend=frontend,
+        on_device=on_device,
+    )
 
 # matmul
 @handle_frontend_method(