Merge branch 'main' into gh/SS-JIA/215/orig

* main:
  [ET-VK] Enable int8 tiled compute shader to be used with buffer tensors (#10415)
  Revert "Migrate elementwise_util callers to the variants with out_dtypes in template arguments" (#10411)
  Revert "Save some size in dtype_util when dtype selective build is not in use" (#10410)
  Generalize view_copy fusion.
  Add job_arn  to benchmark result (#10372)
  Revert "Minibench refactor (#10376)" (#10405)
  Revert "Arm backend: Populate __init__.py for quantizer and Arm root" (#10395)
  Arm backend: Convert assert to throw ValueError in op_log (#10392)
  Arm backend: Add Tutorial to Example tab on the Docs page (#10386)
  Arm backend: Update node visitors to support TOSA 1.0 (#10390)
  Arm backend: Make it easier to generate non delegated/quantized PTEs (#10387)
  Arm backend: Allow --quantize in non delegated using aot_arm_compiler (#10385)
  Arm backend: Set REGIONCFG registers of the Ethos-U (#10388)

Author: SS-JIA

.github/scripts/extract_benchmark_results.py

@@ -341,6 +341,7 @@ def transform(
     benchmark_results: List,
     benchmark_config: Dict[str, str],
     job_name: str,
+    job_report: Any = {},
 ) -> List:
     """
     Transform the benchmark results into the format writable into the benchmark database
@@ -361,6 +362,7 @@ def transform(
                     # Just keep a copy of the benchmark config here
                     "benchmark_config": json.dumps(benchmark_config),
                     "job_conclusion": "SUCCESS",
+                    "job_arn": job_report.get("arn", ""),
                 },
             },
             "model": {
@@ -446,6 +448,7 @@ def transform_failure_record(
                 "app_type": app_type,
                 "job_conclusion": result,
                 "failure_type": level,
+                "job_arn": report.get("arn", ""),
                 "job_report": json.dumps(report),
             },
         },
@@ -512,6 +515,7 @@ def get_benchmark_config(
 def extract_benchmark_result_from_artifact(
     artifact: Dict[str, Any],
     benchmark_config: Dict[str, str],
+    job_report: Any,
 ) -> List[Any]:
     job_name = artifact.get("job_name", "")
     artifact_type = artifact.get("type", "")
@@ -532,7 +536,9 @@ def extract_benchmark_result_from_artifact(
         )
     if not benchmark_results:
         return []
-    return transform(app_type, benchmark_results, benchmark_config, job_name)
+    return transform(
+        app_type, benchmark_results, benchmark_config, job_name, job_report
+    )
 
 
 def get_app_type(type: str):
@@ -674,7 +680,7 @@ def process_benchmark_results(content: Any, app: str, benchmark_configs: str):
             for job_artifact in job_artifacts:
                 # generate result for each schema
                 results = extract_benchmark_result_from_artifact(
-                    job_artifact, benchmark_config
+                    job_artifact, benchmark_config, job_report
                 )
                 all_benchmark_results.extend(results)
     return all_benchmark_results

backends/arm/__init__.py

@@ -14,9 +14,9 @@
 import logging
 from typing import final, List
 
-from executorch.backends.arm import TOSABackend
-
 from executorch.backends.arm.arm_vela import vela_compile
+
+from executorch.backends.arm.tosa_backend import TOSABackend
 from executorch.exir.backend.backend_details import BackendDetails, PreprocessResult
 from executorch.exir.backend.compile_spec_schema import CompileSpec
 from torch.export.exported_program import ExportedProgram

backends/arm/ethosu_backend.py

@@ -10,7 +10,8 @@
 from executorch.backends.arm.arm_backend import (
     is_ethosu,
 )  # usort: skip
-from executorch.backends.arm import EthosUBackend, TOSAPartitioner
+from executorch.backends.arm.ethosu_backend import EthosUBackend
+from executorch.backends.arm.tosa_partitioner import TOSAPartitioner
 from executorch.exir.backend.compile_spec_schema import CompileSpec
 from executorch.exir.backend.partitioner import DelegationSpec
 from torch.fx.passes.operator_support import OperatorSupportBase

backends/arm/ethosu_partitioner.py

@@ -77,7 +77,6 @@ def is_node_tosa_supported(
     ) -> bool:
         assert node.target in self.targets
 
-        assert tosa_spec.support_integer()
         supported_dtypes = (
             self.ALL_SUPPORTED_TYPES
             if tosa_spec.support_float()

backends/arm/operator_support/to_copy_support.py

@@ -2,9 +2,8 @@
 #
 # 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 List
+from typing import Any, List
 
-import tosa_tools.v0_80.serializer.tosa_serializer as ts
 from executorch.backends.arm._passes.arm_pass_utils import get_first_fake_tensor
 from executorch.backends.arm.operators.node_visitor import (
     NodeVisitor,
@@ -15,19 +14,22 @@
 
 
 @register_node_visitor
-class MaxVisitor(NodeVisitor):
+class MaxVisitor_0_80(NodeVisitor):
     target = "aten.amax.default"
 
+    tosa_specs = NodeVisitor.tosa_specs_0_80
+
     def __init__(self, *args):
         super().__init__(*args)
 
     def define_node(
         self,
         node: Node,
-        tosa_graph: ts.TosaSerializer,
+        tosa_graph: Any,
         inputs: List[TosaArg],
         output: TosaArg,
     ) -> None:
+        import tosa_tools.v0_80.serializer.tosa_serializer as ts
 
         input = inputs[0]
         dim = inputs[1].number
@@ -49,3 +51,42 @@ def define_node(
         tosa_graph.addOperator(
             ts.TosaOp.Op().REDUCE_MAX, [input.name], [output.name], attr
         )
+
+
+@register_node_visitor
+class MaxVisitor(NodeVisitor):
+    target = "aten.amax.default"
+
+    tosa_specs = NodeVisitor.tosa_specs_1_00
+
+    def __init__(self, *args):
+        super().__init__(*args)
+
+    def define_node(
+        self,
+        node: Node,
+        tosa_graph: Any,
+        inputs: List[TosaArg],
+        output: TosaArg,
+    ) -> None:
+        import serializer.tosa_serializer as ts
+
+        input = inputs[0]
+        dim = inputs[1].number
+
+        if dim < 0:
+            tensor = get_first_fake_tensor(node)
+            rank = len(tensor.size())
+            dim = rank + dim
+
+        keep_dims = inputs[2].number
+        if not keep_dims:
+            raise RuntimeError(
+                "TOSA only supports keepdims == True; Did you run the convert_minmax pass?"
+            )
+
+        attr = ts.TosaSerializerAttribute()
+        attr.ReduceMaxAttribute(axis=input.dim_order.index(dim), nan_mode=1)
+        tosa_graph.addOperator(
+            ts.TosaOp.Op().REDUCE_MAX, [input.name], [output.name], attr
+        )

backends/arm/operators/op_amax.py

@@ -2,9 +2,8 @@
 #
 # 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 List
+from typing import Any, List
 
-import tosa_tools.v0_80.serializer.tosa_serializer as ts
 from executorch.backends.arm._passes.arm_pass_utils import get_first_fake_tensor
 from executorch.backends.arm.operators.node_visitor import (
     NodeVisitor,
@@ -15,19 +14,22 @@
 
 
 @register_node_visitor
-class MinVisitor(NodeVisitor):
+class MinVisitor_0_80(NodeVisitor):
     target = "aten.amin.default"
 
+    tosa_specs = NodeVisitor.tosa_specs_0_80
+
     def __init__(self, *args):
         super().__init__(*args)
 
     def define_node(
         self,
         node: Node,
-        tosa_graph: ts.TosaSerializer,
+        tosa_graph: Any,
         inputs: List[TosaArg],
         output: TosaArg,
     ) -> None:
+        import tosa_tools.v0_80.serializer.tosa_serializer as ts
 
         input = inputs[0]
         dim = inputs[1].number
@@ -49,3 +51,42 @@ def define_node(
         tosa_graph.addOperator(
             ts.TosaOp.Op().REDUCE_MIN, [input.name], [output.name], attr
         )
+
+
+@register_node_visitor
+class MinVisitor(NodeVisitor):
+    target = "aten.amin.default"
+
+    tosa_specs = NodeVisitor.tosa_specs_1_00
+
+    def __init__(self, *args):
+        super().__init__(*args)
+
+    def define_node(
+        self,
+        node: Node,
+        tosa_graph: Any,
+        inputs: List[TosaArg],
+        output: TosaArg,
+    ) -> None:
+        import serializer.tosa_serializer as ts
+
+        input = inputs[0]
+        dim = inputs[1].number
+
+        if dim < 0:
+            tensor = get_first_fake_tensor(node)
+            rank = len(tensor.size())
+            dim = rank + dim
+
+        keep_dims = inputs[2].number
+        if not keep_dims:
+            raise RuntimeError(
+                "TOSA only supports keepdims == True; Did you run the convert_minmax pass?"
+            )
+
+        attr = ts.TosaSerializerAttribute()
+        attr.ReduceMinAttribute(axis=input.dim_order.index(dim), nan_mode=1)
+        tosa_graph.addOperator(
+            ts.TosaOp.Op().REDUCE_MIN, [input.name], [output.name], attr
+        )