Update base for Update on "[ET-VK] Introduce vTensorPtr to prevent reference invalidation and remove get_val() API"

## Context

Currently when writing operators developers will save a reference to a `vTensor` retrieved from a `ComputeGraph`'s list of `values_` like so:

```
vTensor& vten = graph.get_val(vref).toTensor();
```

However, this is dangerous since if any values are added once the reference has been stored, `values_` which is a `std::vector` may have been resized and therefore have its contents moved, meaning the reference is now invalid.

To protect against this, this changeset introduces the `vTensorPtr` class which is a wrapper around a `vTensor*`.  When constructed, it will increment a counter in the `ComputeGraph` instance, and when destroyed it will decrement the counter. `ComputeGraph` cannot add any values while the counter is not zero.

Since `Value` can be converted to other non-trivial types, this changeset also removes the `get_val` function entirely to guard against unsafe behaviour.

Differential Revision: [D55984187](https://our.internmc.facebook.com/intern/diff/D55984187/)

[ghstack-poisoned]

Author: SS-JIA

CMakeLists.txt

@@ -144,6 +144,8 @@ option(EXECUTORCH_BUILD_COREML "Build the Core ML backend" OFF)
 
 option(EXECUTORCH_BUILD_CUSTOM "Build the custom kernels" OFF)
 
+option(EXECUTORCH_BUILD_CUSTOM_OPS_AOT "Build the custom ops lib for AOT" OFF)
+
 option(EXECUTORCH_BUILD_EXTENSION_DATA_LOADER "Build the Data Loader extension"
        OFF)
 
@@ -185,12 +187,19 @@ cmake_dependent_option(
 cmake_dependent_option(EXECUTORCH_BUILD_CPUINFO "Build cpuinfo library." ON
                        "NOT EXECUTORCH_BUILD_ARM_BAREMETAL" OFF)
 
+if(EXECUTORCH_BUILD_CUSTOM_OPS_AOT)
+  set(EXECUTORCH_BUILD_CUSTOM ON)
+endif()
+
 if(EXECUTORCH_BUILD_CUSTOM)
   set(EXECUTORCH_BUILD_OPTIMIZED ON)
 endif()
 
 if(EXECUTORCH_BUILD_CPUINFO)
   # --- cpuinfo
+  set(ORIGINAL_CMAKE_POSITION_INDEPENDENT_CODE_FLAG
+      ${CMAKE_POSITION_INDEPENDENT_CODE})
+  set(CMAKE_POSITION_INDEPENDENT_CODE ON)
   set(CPUINFO_SOURCE_DIR "backends/xnnpack/third-party/cpuinfo")
   set(CPUINFO_BUILD_TOOLS
       OFF
@@ -212,10 +221,15 @@ if(EXECUTORCH_BUILD_CPUINFO)
       CACHE STRING "")
   set(CLOG_SOURCE_DIR "${CPUINFO_SOURCE_DIR}/deps/clog")
   add_subdirectory("${CPUINFO_SOURCE_DIR}")
+  set(CMAKE_POSITION_INDEPENDENT_CODE
+      ${ORIGINAL_CMAKE_POSITION_INDEPENDENT_CODE_FLAG})
 endif()
 
 if(EXECUTORCH_BUILD_PTHREADPOOL)
   # --- pthreadpool
+  set(ORIGINAL_CMAKE_POSITION_INDEPENDENT_CODE_FLAG
+      ${CMAKE_POSITION_INDEPENDENT_CODE})
+  set(CMAKE_POSITION_INDEPENDENT_CODE ON)
   set(PTHREADPOOL_SOURCE_DIR "backends/xnnpack/third-party/pthreadpool")
   set(PTHREADPOOL_BUILD_TESTS
       OFF
@@ -235,6 +249,8 @@ if(EXECUTORCH_BUILD_PTHREADPOOL)
        CACHE STRING "")
   endif()
   add_subdirectory("${PTHREADPOOL_SOURCE_DIR}")
+  set(CMAKE_POSITION_INDEPENDENT_CODE
+      ${ORIGINAL_CMAKE_POSITION_INDEPENDENT_CODE_FLAG})
 endif()
 
 if(NOT PYTHON_EXECUTABLE)
@@ -546,6 +562,9 @@ if(EXECUTORCH_BUILD_PYBIND)
     list(APPEND _dep_libs custom_ops)
   endif()
 
+  if(EXECUTORCH_BUILD_CUSTOM_OPS_AOT)
+    list(APPEND _dep_libs custom_ops_aot_lib)
+  endif()
   # compile options for pybind
 
   set(_pybind_compile_options -Wno-deprecated-declarations -fPIC -frtti

backends/qualcomm/CMakeLists.txt

@@ -253,7 +253,7 @@ target_link_libraries(qnn_executorch_backend
     qnn_executorch_header
     qnn_schema
     qnn_manager
-    executorch
+    executorch_no_prim_ops
     qcir_utils
 )
 target_link_libraries(utils

backends/xnnpack/CMakeLists.txt

@@ -81,7 +81,7 @@ add_library(xnnpack_backend STATIC ${_xnnpack_backend__srcs})
 target_link_libraries(xnnpack_backend
                       PRIVATE
                       ${xnnpack_third_party}
-                      executorch
+                      executorch_no_prim_ops
                       xnnpack_schema)
 
 target_include_directories(xnnpack_backend

docs/source/build-run-qualcomm-ai-engine-direct-backend.md

@@ -115,6 +115,10 @@ Python APIs on x64 are required to compile models to Qualcomm AI Engine Direct b
 
 ```bash
 cd $EXECUTORCH_ROOT
+# Workaround for fbs files in exir/_serialize
+cp schema/program.fbs exir/_serialize/program.fbs
+cp schema/scalar_type.fbs exir/_serialize/scalar_type.fbs
+
 mkdir build_x86_64
 cd build_x86_64
 cmake .. -DEXECUTORCH_BUILD_QNN=ON -DQNN_SDK_ROOT=${QNN_SDK_ROOT}
@@ -138,8 +142,8 @@ mkdir build_android
 cd build_android
 # build executorch & qnn_executorch_backend
 cmake .. \
-    -DBUCK2=buck2 \
     -DCMAKE_INSTALL_PREFIX=$PWD \
+    -DEXECUTORCH_BUILD_SDK=ON \
     -DEXECUTORCH_BUILD_QNN=ON \
     -DQNN_SDK_ROOT=$QNN_SDK_ROOT \
     -DCMAKE_TOOLCHAIN_FILE=$ANDROID_NDK/build/cmake/android.toolchain.cmake \
@@ -220,6 +224,7 @@ So, we can run `qnn_executor_runner` like
 ```bash
 adb push ./deeplab_v3/dlv3_qnn.pte ${DEVICE_DIR}
 adb push ${EXECUTORCH_ROOT}/build_android/examples/qualcomm/qnn_executor_runner ${DEVICE_DIR}
+adb push ${EXECUTORCH_ROOT}/build_android/lib/libqnn_executorch_backend.so ${DEVICE_DIR}
 adb shell "cd ${DEVICE_DIR} \
            && export LD_LIBRARY_PATH=${DEVICE_DIR} \
            && export ADSP_LIBRARY_PATH=${DEVICE_DIR} \

docs/source/build-run-xtensa.md

@@ -64,7 +64,7 @@ Step 2. Make sure you have completed the ExecuTorch setup tutorials linked to at
 The working tree is:
 
 ```
-examples/xtensa/
+examples/cadence/
 ├── aot
 ├── kernels
 ├── ops
@@ -75,7 +75,7 @@ examples/xtensa/
 
 ***AoT (Ahead-of-Time) Components***:
 
-The AoT folder contains all of the python scripts and functions needed to export the model to an ExecuTorch `.pte` file. In our case, [export_example.py](https://github.com/pytorch/executorch/blob/main/examples/xtensa/aot/export_example.py) is an API that takes a model (nn.Module) and representative inputs and runs it through the quantizer (from [quantizer.py](https://github.com/pytorch/executorch/blob/main/examples/xtensa/aot/quantizer.py)). Then a few compiler passes, also defined in [quantizer.py](https://github.com/pytorch/executorch/blob/main/examples/xtensa/aot/quantizer.py), will replace operators with custom ones that are supported and optimized on the chip. Any operator needed to compute things should be defined in [meta_registrations.py](https://github.com/pytorch/executorch/blob/main/examples/xtensa/aot/meta_registrations.py) and have corresponding implemetations in the other folders.
+The AoT folder contains all of the python scripts and functions needed to export the model to an ExecuTorch `.pte` file. In our case, [export_example.py](https://github.com/pytorch/executorch/blob/main/examples/cadence/aot/export_example.py) is an API that takes a model (nn.Module) and representative inputs and runs it through the quantizer (from [quantizer.py](https://github.com/pytorch/executorch/blob/main/examples/cadence/aot/quantizer.py)). Then a few compiler passes, also defined in [quantizer.py](https://github.com/pytorch/executorch/blob/main/examples/cadence/aot/quantizer.py), will replace operators with custom ones that are supported and optimized on the chip. Any operator needed to compute things should be defined in [meta_registrations.py](https://github.com/pytorch/executorch/blob/main/examples/cadence/aot/meta_registrations.py) and have corresponding implemetations in the other folders.
 
 ***Operators***:
 
@@ -101,14 +101,14 @@ python3 -m examples.portable.scripts.export --model_name="add"
 ***Quantized Operators***:
 
 The other, more complex model are custom operators, including:
-  - a quantized [linear](https://pytorch.org/docs/stable/generated/torch.nn.Linear.html) operation. The model is defined [here](https://github.com/pytorch/executorch/blob/main/examples/xtensa/tests/quantized_linear_example.py#L28). Linear is the backbone of most Automatic Speech Recognition (ASR) models.
-  - a quantized [conv1d](https://pytorch.org/docs/stable/generated/torch.nn.Conv1d.html) operation. The model is defined [here](https://github.com/pytorch/executorch/blob/main/examples/xtensa/tests/quantized_conv1d_example.py#L36). Convolutions are important in wake word and many denoising models.
+  - a quantized [linear](https://pytorch.org/docs/stable/generated/torch.nn.Linear.html) operation. The model is defined [here](https://github.com/pytorch/executorch/blob/main/examples/cadence/tests/quantized_linear_example.py#L28). Linear is the backbone of most Automatic Speech Recognition (ASR) models.
+  - a quantized [conv1d](https://pytorch.org/docs/stable/generated/torch.nn.Conv1d.html) operation. The model is defined [here](https://github.com/pytorch/executorch/blob/main/examples/cadence/tests/quantized_conv1d_example.py#L36). Convolutions are important in wake word and many denoising models.
 
 In both cases the generated file is called `XtensaDemoModel.pte`.
 
 ```bash
 cd executorch
-python3 -m examples.xtensa.tests.quantized_<linear,conv1d>_example
+python3 -m examples.cadence.tests.quantized_<linear,conv1d>_example
 ```
 
 ***Small Model: RNNT predictor***:
@@ -118,7 +118,7 @@ The predictor is a sequence of basic ops (embedding, ReLU, linear, layer norm) a
 
 ```bash
 cd executorch
-python3 -m examples.xtensa.tests.rnnt_predictor_quantized_example
+python3 -m examples.cadence.tests.rnnt_predictor_quantized_example
 ```
 
 The generated file is called `XtensaDemoModel.pte`.
@@ -131,7 +131,7 @@ In this step, you'll be building the DSP firmware image that consists of the sam
 ***Step 1***. Configure the environment variables needed to point to the Xtensa toolchain that you have installed in the previous step. The three environment variables that need to be set include:
 ```bash
 # Directory in which the Xtensa toolchain was installed
-export XTENSA_TOOLCHAIN=/home/user_name/xtensa/XtDevTools/install/tools
+export XTENSA_TOOLCHAIN=/home/user_name/cadence/XtDevTools/install/tools
 # The version of the toolchain that was installed. This is essentially the name of the directory
 # that is present in the XTENSA_TOOLCHAIN directory from above.
 export TOOLCHAIN_VER=RI-2021.8-linux
@@ -151,7 +151,7 @@ cd executorch
 rm -rf cmake-out
 # prebuild and install executorch library
 cmake -DBUCK2=buck2 \
-    -DCMAKE_TOOLCHAIN_FILE=<path_to_executorch>/examples/xtensa/xtensa.cmake \
+    -DCMAKE_TOOLCHAIN_FILE=<path_to_executorch>/examples/cadence/cadence.cmake \
     -DCMAKE_INSTALL_PREFIX=cmake-out \
     -DCMAKE_BUILD_TYPE=Debug \
     -DPYTHON_EXECUTABLE=python3 \
@@ -165,18 +165,18 @@ cmake -DBUCK2=buck2 \
     -Bcmake-out .
 
 cmake --build cmake-out -j8 --target install --config Debug
-# build xtensa runner
+# build cadence runner
 cmake -DCMAKE_BUILD_TYPE=Debug \
-    -DCMAKE_TOOLCHAIN_FILE=<path_to_executorch>/examples/xtensa/xtensa.cmake \
+    -DCMAKE_TOOLCHAIN_FILE=<path_to_executorch>/examples/cadence/cadence.cmake \
     -DCMAKE_PREFIX_PATH=<path_to_executorch>/cmake-out \
     -DMODEL_PATH=<path_to_program_file_generated_in_previous_step> \
     -DNXP_SDK_ROOT_DIR=<path_to_nxp_sdk_root> -DEXECUTORCH_BUILD_FLATC=0 \
     -DFLATC_EXECUTABLE="$(which flatc)" \
     -DNN_LIB_BASE_DIR=<path_to_nnlib_cloned_in_step_2> \
-    -Bcmake-out/examples/xtensa \
-    examples/xtensa
+    -Bcmake-out/examples/cadence \
+    examples/cadence
 
-cmake --build cmake-out/examples/xtensa -j8 -t xtensa_executorch_example
+cmake --build cmake-out/examples/cadence -j8 -t cadence_executorch_example
 ```
 
 After having succesfully run the above step you should see two binary files in their CMake output directory.
@@ -213,6 +213,6 @@ First 20 elements of output 0
 
 In this tutorial, you have learned how to export a quantized operation, build the ExecuTorch runtime and run this model on the Xtensa HiFi4 DSP chip.
 
-The (quantized linear) model in this tutorial is a typical operation appearing in ASR models, and can be extended to a complete ASR model by creating the model as a new test and adding the needed operators/kernels to [operators](https://github.com/pytorch/executorch/blob/main/examples/xtensa/ops) and [kernels](https://github.com/pytorch/executorch/blob/main/examples/xtensa/kernels).
+The (quantized linear) model in this tutorial is a typical operation appearing in ASR models, and can be extended to a complete ASR model by creating the model as a new test and adding the needed operators/kernels to [operators](https://github.com/pytorch/executorch/blob/main/examples/cadence/ops) and [kernels](https://github.com/pytorch/executorch/blob/main/examples/cadence/kernels).
 
 Other models can be created following the same structure, always assuming that operators and kernels are available.

examples/xtensa/CMakeLists.txt renamed to examples/cadence/CMakeLists.txt

@@ -12,7 +12,7 @@ if(NOT CMAKE_CXX_STANDARD)
 endif()
 
 # Set the project name.
-project(xtensa_executorch_example)
+project(cadence_executorch_example)
 
 # Source root directory for executorch.
 if(NOT EXECUTORCH_ROOT)
@@ -100,21 +100,21 @@ add_custom_command(
 
 add_custom_target(gen_model_header DEPENDS ${CMAKE_BINARY_DIR}/model_pte.h)
 
-add_executable(xtensa_executorch_example executor_runner.cpp)
-add_dependencies(xtensa_executorch_example gen_model_header)
+add_executable(cadence_executorch_example executor_runner.cpp)
+add_dependencies(cadence_executorch_example gen_model_header)
 
 # lint_cmake: -linelength
-target_include_directories(xtensa_executorch_example PUBLIC ${ROOT_DIR}/..
+target_include_directories(cadence_executorch_example PUBLIC ${ROOT_DIR}/..
                                                             ${CMAKE_BINARY_DIR}
                                                             ${_common_include_directories})
 
-target_link_options(xtensa_executorch_example PRIVATE
+target_link_options(cadence_executorch_example PRIVATE
                     -mlsp=${NXP_SDK_ROOT_DIR}/devices/MIMXRT685S/xtensa/min-rt)
-target_link_libraries(xtensa_executorch_example dsp_mu_polling_libs
-                      xtensa_ops_lib extension_runner_util executorch)
+target_link_libraries(cadence_executorch_example dsp_mu_polling_libs
+                      cadence_ops_lib extension_runner_util executorch)
 
 add_custom_command(
-  TARGET xtensa_executorch_example
+  TARGET cadence_executorch_example
   POST_BUILD
   COMMAND
     ${PYTHON_EXECUTABLE} ${CMAKE_CURRENT_LIST_DIR}/utils/post_compilation.py

examples/xtensa/aot/compiler.py renamed to examples/cadence/aot/compiler.py

@@ -17,20 +17,20 @@
 
 from .compiler import export_to_edge
 from .quantizer import (
+    CadenceBaseQuantizer,
     QuantFusion,
-    ReplacePT2DequantWithXtensaDequant,
-    ReplacePT2QuantWithXtensaQuant,
-    XtensaBaseQuantizer,
+    ReplacePT2DequantWithCadenceDequant,
+    ReplacePT2QuantWithCadenceQuant,
 )
 
 
 FORMAT = "[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s"
 logging.basicConfig(level=logging.INFO, format=FORMAT)
 
 
-def export_xtensa_model(model, example_inputs):
+def export_model(model, example_inputs):
     # Quantizer
-    quantizer = XtensaBaseQuantizer()
+    quantizer = CadenceBaseQuantizer()
 
     # Export
     model_exp = capture_pre_autograd_graph(model, example_inputs)
@@ -42,24 +42,24 @@ def export_xtensa_model(model, example_inputs):
     # Convert
     converted_model = convert_pt2e(prepared_model)
 
-    # pyre-fixme[16]: Pyre doesn't get that XtensaQuantizer has a patterns attribute
+    # pyre-fixme[16]: Pyre doesn't get that CadenceQuantizer has a patterns attribute
     patterns = [q.pattern for q in quantizer.quantizers]
     QuantFusion(patterns)(converted_model)
 
-    # Get edge program (note: the name will change to export_to_xtensa in future PRs)
+    # Get edge program (note: the name will change to export_to_cadence in future PRs)
     edge_prog_manager = export_to_edge(converted_model, example_inputs, pt2_quant=True)
 
     # Run a couple required passes for quant/dequant ops
-    xtensa_prog_manager = edge_prog_manager.transform(
-        [ReplacePT2QuantWithXtensaQuant(), ReplacePT2DequantWithXtensaDequant()],
+    cadence_prog_manager = edge_prog_manager.transform(
+        [ReplacePT2QuantWithCadenceQuant(), ReplacePT2DequantWithCadenceDequant()],
         check_ir_validity=False,
     )
 
-    exec_prog = xtensa_prog_manager.to_executorch()
+    exec_prog = cadence_prog_manager.to_executorch()
 
     logging.info(
         f"Final exported graph module:\n{exec_prog.exported_program().graph_module}"
     )
 
-    # Save the program as XtensaDemoModel.pte
-    save_pte_program(exec_prog, "XtensaDemoModel")
+    # Save the program as CadenceDemoModel.pte
+    save_pte_program(exec_prog, "CadenceDemoModel")

examples/xtensa/aot/export_example.py renamed to examples/cadence/aot/export_example.py

@@ -12,7 +12,7 @@
 
 from .utils import get_conv1d_output_size
 
-lib = Library("xtensa", "DEF")
+lib = Library("cadence", "DEF")
 
 lib.define(
     "quantize_per_tensor(Tensor input, float scale, int zero_point, int quant_min, int quant_max, ScalarType dtype) -> (Tensor Z)"
@@ -56,7 +56,7 @@
     "quantized_conv.out(Tensor input, Tensor weight, Tensor bias, int[] stride, SymInt[] padding, int[] dilation, int groups, int input_zero_point, Tensor weight_zero_point, Tensor bias_scale, float out_scale, int out_zero_point, Tensor out_multiplier, Tensor out_shift, bool channel_last=False, *, Tensor(a!) out) -> Tensor(a!)"
 )
 
-m = Library("xtensa", "IMPL", "Meta")
+m = Library("cadence", "IMPL", "Meta")
 
 
 @impl(m, "quantize_per_tensor")