Fix the build of examples/xtensa (#2642)

Summary:
Pull Request resolved: #2642

Fixes to the aot side and the build side following the new docs.

bypass-github-pytorch-ci-checks

Reviewed By: tarun292, cccclai

Differential Revision: D55300249

fbshipit-source-id: e3f4b456ed4fada037fc9c0f074601a043240872

docs/source/build-run-xtensa.md

@@ -136,15 +136,22 @@ 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_INSTALL_PREFIX=cmake-out \
     -DCMAKE_BUILD_TYPE=Debug \
+    -DEXECUTORCH_BUILD_HOST_TARGETS=ON \
+    -DEXECUTORCH_BUILD_EXECUTOR_RUNNER=OFF \
+    -DEXECUTORCH_BUILD_FLATC=OFF \
+    -DFLATC_EXECUTABLE="$(which flatc)" \
+    -DEXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL=ON \
     -DPYTHON_EXECUTABLE=python3 \
     -Bcmake-out .
 
 cmake --build cmake-out -j8 --target install --config Debug
 # build xtensa runner
 cmake -DCMAKE_BUILD_TYPE=Debug \
-    -DCMAKE_TOOLCHAIN_FILE=../examples/xtensa/xtensa.cmake \
+    -DCMAKE_TOOLCHAIN_FILE=<path_to_executorch>/examples/xtensa/xtensa.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)" \

examples/xtensa/CMakeLists.txt

@@ -11,6 +11,10 @@ if(NOT CMAKE_CXX_STANDARD)
   set(CMAKE_CXX_STANDARD 17)
 endif()
 
+if(NOT PYTHON_EXECUTABLE)
+  set(PYTHON_EXECUTABLE python3)
+endif()
+
 # Set the project name.
 project(xtensa_executorch_example)
 
@@ -19,6 +23,9 @@ if(NOT EXECUTORCH_ROOT)
   set(EXECUTORCH_ROOT ${CMAKE_CURRENT_SOURCE_DIR}/../..)
 endif()
 
+# Let files say "include <executorch/path/to/header.h>".
+set(_common_include_directories ${EXECUTORCH_ROOT}/..)
+
 # Find prebuilt executorch lib
 find_package(executorch CONFIG REQUIRED)
 
@@ -74,6 +81,10 @@ target_include_directories(
          ${NXP_SDK_ROOT_DIR}/devices/MIMXRT685S/utilities/str
          ${NXP_SDK_ROOT_DIR}/boards/evkmimxrt685/dsp_examples/mu_polling/dsp)
 
+add_library(extension_runner_util STATIC IMPORTED)
+set_property(TARGET extension_runner_util PROPERTY IMPORTED_LOCATION
+  "${CMAKE_CURRENT_LIST_DIR}/../../cmake-out/extension/runner_util/libextension_runner_util.a")
+
 add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/ops)
 add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/kernels)
 
@@ -90,13 +101,15 @@ 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)
 
+# lint_cmake: -linelength
 target_include_directories(xtensa_executorch_example PUBLIC ${ROOT_DIR}/..
-                                                            ${CMAKE_BINARY_DIR})
+                                                            ${CMAKE_BINARY_DIR}
+                                                            ${_common_include_directories})
 
 target_link_options(xtensa_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 executorch)
+                      xtensa_ops_lib extension_runner_util executorch)
 
 add_custom_command(
   TARGET xtensa_executorch_example

examples/xtensa/aot/export_example.py

@@ -11,7 +11,7 @@
 from .meta_registrations import *  # noqa
 
 import torch
-from executorch.exir import EdgeCompileConfig, ExecutorchBackendConfig
+from executorch.exir import EdgeCompileConfig
 from torch._export import capture_pre_autograd_graph
 from torch.ao.quantization.quantize_pt2e import convert_pt2e, prepare_pt2e
 
@@ -77,14 +77,15 @@ def forward(self, x: torch.Tensor):
         export_to_edge(
             converted_model_exp,
             example_inputs,
-            EdgeCompileConfig(
+            edge_compile_config=EdgeCompileConfig(
                 _check_ir_validity=False,
             ),
         )
         .transform(
-            [ReplacePT2QuantWithXtensaQuant(), ReplacePT2DequantWithXtensaDequant()]
+            [ReplacePT2QuantWithXtensaQuant(), ReplacePT2DequantWithXtensaDequant()],
+            check_ir_validity=False,
         )
-        .to_executorch(config=ExecutorchBackendConfig(extract_constant_segment=False))
+        .to_executorch()
     )
 
     logging.info(f"Final exported graph:\n{exec_prog.exported_program().graph}")

examples/xtensa/executor_runner.cpp

@@ -27,12 +27,12 @@
 #include <vector>
 
 #include <executorch/extension/data_loader/buffer_data_loader.h>
+#include <executorch/extension/runner_util/inputs.h>
 #include <executorch/runtime/executor/method.h>
 #include <executorch/runtime/executor/program.h>
 #include <executorch/runtime/platform/log.h>
 #include <executorch/runtime/platform/profiler.h>
 #include <executorch/runtime/platform/runtime.h>
-#include <executorch/util/util.h>
 
 static uint8_t method_allocator_pool[18 * 1024U]; // 4 MB
 
@@ -177,7 +177,7 @@ int main(int argc, char** argv) {
   }
 
   ET_LOG(Info, "Method loaded.");
-  torch::executor::util::PrepareInputTensors(*method);
+  torch::executor::util::prepare_input_tensors(*method);
   ET_LOG(Info, "Starting the model execution...");
 
   Error status = method->execute();

examples/xtensa/ops/CMakeLists.txt

@@ -32,10 +32,21 @@ set(_aten_ops__srcs
 add_library(aten_ops_xtensa ${_aten_ops__srcs})
 target_link_libraries(aten_ops_xtensa PUBLIC executorch)
 
+# Let files say "include <executorch/path/to/header.h>".
+set(_common_include_directories ${EXECUTORCH_ROOT}/..)
+
+target_include_directories(aten_ops_xtensa PUBLIC ${ROOT_DIR}/..
+                                                  ${CMAKE_BINARY_DIR}
+                                                  ${_common_include_directories})
+
 # Custom ops that are needed to run the test model.
 add_library(
   custom_ops "quantized_linear_out.cpp" "quantize_per_tensor.cpp"
   "dequantize_per_tensor.cpp")
+target_include_directories(custom_ops PUBLIC ${ROOT_DIR}/..
+                                             ${CMAKE_BINARY_DIR}
+                                             ${_common_include_directories})
+
 target_link_libraries(custom_ops PUBLIC executorch)
 target_link_libraries(custom_ops PRIVATE xtensa_kernels)
 

examples/xtensa/ops/dequantize_per_tensor.cpp

@@ -25,15 +25,7 @@ void dequantize_per_tensor_out(
     int64_t quant_min,
     int64_t quant_max,
     ScalarType dtype,
-    exec_aten::optional<ScalarType>& out_dtype,
     Tensor& out) {
-  if (out_dtype.has_value()) {
-    ET_CHECK_MSG(
-        out_dtype.value() == ScalarType::Float,
-        "Expected out dtype to be Float but got %hhd",
-        out_dtype.value());
-  }
-
   float* out_data = out.mutable_data_ptr<float>();
   size_t numel = out.numel();
 

examples/xtensa/ops/functions.yaml

@@ -26,7 +26,7 @@
     - arg_meta: null
       kernel_name: impl::HiFi::quantized_linear_pt2_out
 
-- func: xtensa::dequantize_per_tensor.out(Tensor input, float scale, int zero_point, int quant_min, int quant_max, ScalarType dtype, *, ScalarType? out_dtype=None, Tensor(a!) out) -> Tensor(a!)
+- func: xtensa::dequantize_per_tensor.out(Tensor input, float scale, int zero_point, int quant_min, int quant_max, ScalarType dtype, *, Tensor(a!) out) -> Tensor(a!)
   variants: function
   kernels:
     - arg_meta: null