Update on "[ET-VK] Fix SDK event_name for inputs and outputs"

## Issue
In the ET-SDK, we assign an `event_name` to each operation. In ET-VK, we compose a unique `event_name` using the `node_id`. The `node_id` exists for every `OperatorCall` but not for input/output with `nchw_to_image`/`image_to_nchw`. Those cases collapse into `node_id == 0` which means all `nchw_to_image` had the same `event_name` and hence only one was stored. The same reasoning results in storage of only one `image_to_nchw`.

## Solution
Ignore the serialized `node_id` and use the operation's `prepack_node`/`execute_node` vector index.

TODO: Determine if we can remove the serialized `node_id`, or whether this should be fixed differently and still reference it.

Differential Revision: [D68344534](https://our.internmc.facebook.com/intern/diff/D68344534/)

[ghstack-poisoned]

Author: jorgep31415

.ci/scripts/utils.sh

@@ -17,17 +17,17 @@ retry () {
 }
 
 clean_executorch_install_folders() {
-  ./install_requirements.sh --clean
+  ./install_executorch.sh --clean
 }
 
 install_executorch() {
   which pip
   # Install executorch, this assumes that Executorch is checked out in the
   # current directory.
   if [[ "${1:-}" == "use-pt-pinned-commit" ]]; then
-    ./install_requirements.sh --pybind xnnpack --use-pt-pinned-commit
+    ./install_executorch.sh --pybind xnnpack --use-pt-pinned-commit
   else
-    ./install_requirements.sh --pybind xnnpack
+    ./install_executorch.sh --pybind xnnpack
   fi
   # Just print out the list of packages for debugging
   pip list

.github/workflows/apple.yml

@@ -9,7 +9,7 @@ on:
     paths:
       - .ci/scripts/setup-ios.sh
       - .github/workflows/apple.yml
-      - install_requirements.sh
+      - install_executorch.sh
       - backends/apple/**
       - build/build_apple_frameworks.sh
       - build/build_apple_llm_demo.sh

.github/workflows/pull.yml

@@ -200,7 +200,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "cmake"
 
         # install pybind
-        bash install_requirements.sh --pybind xnnpack
+        bash install_executorch.sh --pybind xnnpack
 
         # install Llava requirements
         bash examples/models/llama/install_requirements.sh
@@ -436,7 +436,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "cmake"
 
         # install pybind
-        bash install_requirements.sh --pybind xnnpack
+        bash install_executorch.sh --pybind xnnpack
 
         # install phi-3-mini requirements
         bash examples/models/phi-3-mini/install_requirements.sh
@@ -463,7 +463,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "cmake"
 
         # install pybind
-        bash install_requirements.sh --pybind xnnpack
+        bash install_executorch.sh --pybind xnnpack
 
         # install llama requirements
         bash examples/models/llama/install_requirements.sh
@@ -490,7 +490,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "cmake"
 
         # install pybind
-        bash install_requirements.sh --pybind xnnpack
+        bash install_executorch.sh --pybind xnnpack
 
         # install llama requirements
         bash examples/models/llama/install_requirements.sh
@@ -517,7 +517,7 @@ jobs:
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "cmake"
 
         # install pybind
-        bash install_requirements.sh --pybind xnnpack
+        bash install_executorch.sh --pybind xnnpack
 
         # install llama requirements
         bash examples/models/llama/install_requirements.sh

backends/apple/mps/setup.md

@@ -97,7 +97,7 @@ I 00:00:00.122615 executorch:mps_executor_runner.mm:501] Model verified successf
 ### [Optional] Run the generated model directly using pybind
 1. Make sure `pybind` MPS support was installed:
 ```bash
-./install_requirements.sh --pybind mps
+./install_executorch.sh --pybind mps
 ```
 2. Run the `mps_example` script to trace the model and run it directly from python:
 ```bash

backends/arm/_passes/annotate_channels_last_dim_order_pass.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Arm Limited and/or its affiliates.
+# Copyright 2024-2025 Arm Limited and/or its affiliates.
 # All rights reserved.
 #
 # This source code is licensed under the BSD-style license found in the
@@ -15,7 +15,7 @@
     get_node_arg,
     insert_q_dq_pair,
 )
-from executorch.backends.arm.tosa_quant_utils import dq_op, q_op, register_passable_op
+from executorch.backends.arm.tosa_quant_utils import dq_op, q_op
 from executorch.backends.arm.tosa_utils import is_consumer_node_depthwise_conv2d
 from executorch.exir.dialects._ops import ops as exir_ops
 from executorch.exir.pass_base import ExportPass, PassResult
@@ -43,9 +43,6 @@ def _transpose_impl(*args, **kwargs):
     return args[0]
 
 
-register_passable_op(torch.ops.passthrough_to_tosa._transpose)
-
-
 class AnnotateChannelsLastDimOrder(ExportPass):
     """
     Annotates each node with a tosa_dim_order. tosa_dim_order can be seen as a channels-last dim-order

backends/arm/operators/__init__.py

@@ -13,7 +13,6 @@
     op_bmm,
     op_cat,
     op_conv2d,
-    op_dequant,
     op_exp,
     op_full,
     op_get_item,
@@ -24,7 +23,6 @@
     op_min,
     op_mul,
     op_permute,
-    op_quant,
     op_reciprocal,
     op_relu,
     op_repeat,

backends/arm/operators/op_dequant.py

@@ -1,4 +1,4 @@
-# Copyright 2023-2024 Arm Limited and/or its affiliates.
+# Copyright 2023-2025 Arm Limited and/or its affiliates.
 #
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
@@ -19,7 +19,6 @@
 )
 from executorch.backends.arm.tosa_mapping import TosaArg
 
-from executorch.backends.arm.tosa_quant_utils import quantize_value
 from serializer.tosa_serializer import TosaOp
 
 
@@ -44,8 +43,8 @@ def define_node(
             input_qparams = get_input_qparams(node)  # pyre-ignore[16]
             qargs = input_qparams[0]
             # Convert to quantized representation
-            clamp_min_qs = quantize_value(inputs[1].number, qargs)
-            clamp_max_qs = quantize_value(inputs[2].number, qargs)
+            clamp_min_qs = qargs.quantize_value(inputs[1].number).item()
+            clamp_max_qs = qargs.quantize_value(inputs[2].number).item()
             # Set fp values to 0.0 since they are not used
             clamp_min_fp = 0.0
             clamp_max_fp = 0.0

backends/arm/operators/op_hardtanh.py

@@ -1,11 +1,10 @@
-# Copyright 2024 Arm Limited and/or its affiliates.
+# Copyright 2024-2025 Arm Limited and/or its affiliates.
 #
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
 # pyre-unsafe
 
-import executorch.backends.arm.tosa_quant_utils as tqutils
 import serializer.tosa_serializer as ts
 import torch.fx
 
@@ -43,9 +42,8 @@ def define_node(
         clamp_max_qs = 0
         if inputs[0].dtype == ts.DType.INT8:
             out_qargs = get_output_qparams(node)  # pyre-ignore[16]
-            clamp_min_qs = tqutils.quantize_value(0, out_qargs[0])
-            clamp_max_qs = tqutils.quantize_value(float("inf"), out_qargs[0])
-
+            clamp_min_qs = out_qargs[0].quantize_value(0).item()
+            clamp_max_qs = out_qargs[0].quantize_value(float("inf")).item()
         else:
             clamp_min_fp = 0
             clamp_max_fp = float("inf")