Add to.dtype and neg ops (#8041)

Summary:

#8041
add neg ops and annonation of to.dtype

Differential Revision: D68815927

Author: billmguo

backends/qualcomm/_passes/layout_transform.py

@@ -74,6 +74,7 @@ class LayoutTransform(ExportPass):
         exir_ops.edge.aten.mean.dim,
         exir_ops.edge.aten.minimum.default,
         exir_ops.edge.aten.mul.Tensor,
+        exir_ops.edge.aten.neg.default,
         exir_ops.edge.aten.pow.Tensor_Scalar,
         exir_ops.edge.aten.prelu.default,
         exir_ops.edge.aten.repeat.default,
@@ -86,6 +87,7 @@ class LayoutTransform(ExportPass):
         exir_ops.edge.aten.sum.dim_IntList,
         exir_ops.edge.aten.topk.default,
         exir_ops.edge.aten._to_copy.default,
+        exir_ops.edge.aten.to.dtype,
         *q_ops,
         *dq_ops,
         _operator.getitem,

backends/qualcomm/builders/__init__.py

@@ -46,6 +46,7 @@
     op_mean_dim,
     op_min,
     op_mul,
+    op_neg,
     op_pad,
     op_pow,
     op_prelu,
@@ -118,6 +119,7 @@
     op_mean_dim,
     op_min,
     op_mul,
+    op_neg,
     op_pad,
     op_pow,
     op_prelu,

backends/qualcomm/builders/op_neg.py

@@ -0,0 +1,53 @@
+# Copyright (c) Qualcomm Innovation Center, Inc.
+# All rights reserved
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+from typing import Dict
+
+import executorch.backends.qualcomm.python.PyQnnWrapperAdaptor as PyQnnWrapper
+import torch
+
+from .node_visitor import NodeVisitor, register_node_visitor
+from .qnn_constants import OpElementWiseNeg, QNN_OP_PACKAGE_NAME_QTI_AISW
+
+
+@register_node_visitor
+class Neg(NodeVisitor):
+    target = ["aten.neg.default"]
+
+    def __init__(self, *args) -> None:
+        super().__init__(*args)
+
+    def define_node(
+        self,
+        node: torch.fx.Node,
+        nodes_to_wrappers: Dict[torch.fx.Node, PyQnnWrapper.TensorWrapper],
+    ) -> PyQnnWrapper.PyQnnOpWrapper:
+        input_node = node.args[0]
+        input_tensor = self.get_tensor(input_node, node)
+        neg_inp_tensor_wrapper = self.define_tensor(
+            input_node,
+            node,
+            input_tensor,
+            PyQnnWrapper.Qnn_TensorType_t.QNN_TENSOR_TYPE_NATIVE,
+            nodes_to_wrappers,
+        )
+        neg_input_tensors = [neg_inp_tensor_wrapper]
+        output_tensor = self.get_tensor(node, node)
+        output_tensor_wrapper = self.define_tensor(
+            node,
+            node,
+            output_tensor,
+            PyQnnWrapper.Qnn_TensorType_t.QNN_TENSOR_TYPE_NATIVE,
+            nodes_to_wrappers,
+        )
+        neg_output_tensors = [output_tensor_wrapper]
+        neg_op = PyQnnWrapper.PyQnnOpWrapper(
+            node.name,
+            QNN_OP_PACKAGE_NAME_QTI_AISW,
+            OpElementWiseNeg.op_name,
+        )
+        neg_op.AddInputTensors(neg_input_tensors)
+        neg_op.AddOutputTensors(neg_output_tensors)
+        return neg_op

backends/qualcomm/builders/op_to.py

@@ -16,7 +16,11 @@
 
 @register_node_visitor
 class To(NodeVisitor):
-    target = ["aten._to_copy.default", "dim_order_ops._to_dim_order_copy.default"]
+    target = [
+        "aten._to_copy.default",
+        "dim_order_ops._to_dim_order_copy.default",
+        "aten.to.dtype",
+    ]
     sufixed_8_offset_diff = 128
     sufixed_16_offset_diff = 32768
     epsilon = 1e-6

backends/qualcomm/builders/qnn_constants.py

@@ -145,6 +145,11 @@ class OpElementWiseMultiply:
     op_name: str = "ElementWiseMultiply"
 
 
+@dataclass(init=False, frozen=True)
+class OpElementWiseNeg:
+    op_name: str = "ElementWiseNeg"
+
+
 @dataclass(init=False, frozen=True)
 class OpElementWiseNeuron:
     op_name: str = "ElementWiseNeuron"

backends/qualcomm/quantizer/annotators.py

@@ -396,6 +396,11 @@ def annotate_max_pool2d_with_indices(
     annotate_single_in_single_out(node, quantization_config)
 
 
+@register_annotator([torch.ops.aten.neg.default])
+def annotate_neg(node: Node, quantization_config: QuantizationConfig) -> None:
+    annotate_single_in_single_out(node, quantization_config)
+
+
 @register_annotator([torch.ops.aten.adaptive_avg_pool2d.default])
 def annotate_adaptive_avgpool2d(
     node: Node, quantization_config: QuantizationConfig

backends/qualcomm/tests/models.py

@@ -882,6 +882,14 @@ def forward(self, x):
         return attn_output
 
 
+class Neg(torch.nn.Module):
+    def __init__(self):
+        super().__init__()
+
+    def forward(self, x):
+        return torch.neg(x)
+
+
 class Pad(torch.nn.Module):
     def __init__(self):
         super().__init__()

backends/qualcomm/tests/test_qnn_delegate.py

@@ -540,6 +540,11 @@ def test_qnn_backend_minimum(self):
         sample_input = (torch.randn(1, 2, 3, 4), torch.randn(2, 3, 4))
         self.lower_module_and_test_output(module, sample_input)
 
+    def test_qnn_backend_neg(self):
+        module = Neg()  # noqa: F405
+        sample_input = (torch.randn(1, 4, 16, 16),)
+        self.lower_module_and_test_output(module, sample_input)
+
     def test_qnn_backend_pad(self):
         module = Pad()  # noqa: F405
         sample_input = (torch.randn([1, 8, 128]),)