Ci

.github/workflows/develop.yaml

@@ -32,11 +32,21 @@ jobs:
         uses: actions/setup-python@v5
         with:
           python-version: "3.12"
+          cache: pip
           token: ${{ secrets.GITHUB_TOKEN }}
+      - name: Install Python requirements
+        run: pip install -r scripts/requirements.txt
 
       ############ Install executorch deps ############
+      - name: Cache Cpp ExecuTorch compiled libraries
+        id: cache-executorch-cpp-libs
+        uses: actions/cache@v4
+        with:
+          path: executorch-sys/third-party/executorch/cmake-out
+          key: cache-executorch-cpp-libs-${{ runner.OS }}-${{ hashFiles('scripts/setup_dev.py') }}
       - name: Setup dev env, install executorch and its deps
-        run: python scripts/setup_dev.py
+        if: steps.cache-executorch-cpp-libs.outputs.cache-hit != 'true'
+        run: python scripts/setup_dev.py --skip-executorch-python
 
       ############ Build executorch-rs ############
       - name: Build
@@ -53,9 +63,11 @@ jobs:
         run: cargo test --workspace --all-features
 
       ############ Examples ############
+      - name: Run 'hello world add no_std' example
+        # python export_model.py
+        run: cargo run
+        working-directory: examples/hello_world_add_no_std
       # - name: Run 'hello world add' example
+      #   # python export_model.py
       #   run: cargo run
       #   working-directory: examples/hello_world_add
-      # - name: Run 'hello world add no_std' example
-      #   run: cargo run
-      #   working-directory: examples/hello_world_add_no_std

README.md

@@ -114,22 +114,22 @@ Additional libs are required if feature flags are enabled (see next section):
 The static libraries of the kernels implementations are required only if your model uses them, and they should be **linked manually** by the binary that uses the `executorch` crate.
 For example, the `hello_world_add` example uses a model with a single addition operation, so it compile the C++ library with `DEXECUTORCH_SELECT_OPS_LIST=aten::add.out` and contain the following lines in its `build.rs`:
 ```rust
-println!("cargo::rustc-link-lib=static=portable_kernels");
+println!("cargo::rustc-link-lib=static:+whole-archive=portable_kernels");
 println!("cargo::rustc-link-lib=static:+whole-archive=portable_ops_lib");
 
 let libs_dir = std::env::var("EXECUTORCH_RS_EXECUTORCH_LIB_DIR").unwrap();
 println!("cargo::rustc-link-search={}/kernels/portable/", libs_dir);
 ```
-Note that the `portable_ops_lib` is linked with `+whole-archive` to ensure that all symbols are included in the binary.
+Note that the ops and kernels libs are linked with `+whole-archive` to ensure that all symbols are included in the binary.
 
 ## Cargo Features
 - `data-loader`
 
-    include the `FileDataLoader` struct. The `libextension_data_loader.a` static library is required, compile C++ `executorch` with `EXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON`.
+    Includes the `FileDataLoader` and `MmapDataLoader` structs. Without this feature the only available data loader is `BufferDataLoader`. The `libextension_data_loader.a` static library is required, compile C++ `executorch` with `EXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON`.
 
 - `module`
 
-    include the `Module` struct. The `libextension_module_static.a` static library is required, compile C++ `executorch` with `EXECUTORCH_BUILD_EXTENSION_MODULE=ON`.
+    Includes the `Module` struct. The `libextension_module_static.a` static library is required, compile C++ `executorch` with `EXECUTORCH_BUILD_EXTENSION_MODULE=ON`.
     Also includes the `std` feature.
 
 - `f16`

examples/hello_world_add/build.rs

@@ -1,5 +1,5 @@
 fn main() {
-    println!("cargo::rustc-link-lib=static=portable_kernels");
+    println!("cargo::rustc-link-lib=static:+whole-archive=portable_kernels");
     println!("cargo::rustc-link-lib=static:+whole-archive=portable_ops_lib");
 
     println!("cargo::rerun-if-env-changed=EXECUTORCH_RS_EXECUTORCH_LIB_DIR");

examples/hello_world_add_no_std/build.rs

@@ -1,5 +1,5 @@
 fn main() {
-    println!("cargo::rustc-link-lib=static=portable_kernels");
+    println!("cargo::rustc-link-lib=static:+whole-archive=portable_kernels");
     println!("cargo::rustc-link-lib=static:+whole-archive=portable_ops_lib");
 
     println!("cargo::rerun-if-env-changed=EXECUTORCH_RS_EXECUTORCH_LIB_DIR");

executorch-sys/README.md

@@ -62,22 +62,22 @@ Additional libs are required if feature flags are enabled (see next section):
 The static libraries of the kernels implementations are required only if your model uses them, and they should be **linked manually** by the binary that uses the `executorch` crate.
 For example, the `hello_world_add` example uses a model with a single addition operation, so it compile the C++ library with `DEXECUTORCH_SELECT_OPS_LIST=aten::add.out` and contain the following lines in its `build.rs`:
 ```rust
-println!("cargo::rustc-link-lib=static=portable_kernels");
+println!("cargo::rustc-link-lib=static:+whole-archive=portable_kernels");
 println!("cargo::rustc-link-lib=static:+whole-archive=portable_ops_lib");
 
 let libs_dir = std::env::var("EXECUTORCH_RS_EXECUTORCH_LIB_DIR").unwrap();
 println!("cargo::rustc-link-search={}/kernels/portable/", libs_dir);
 ```
-Note that the `portable_ops_lib` is linked with `+whole-archive` to ensure that all symbols are included in the binary.
+Note that the ops and kernels libs are linked with `+whole-archive` to ensure that all symbols are included in the binary.
 
 ## Cargo Features
 - `data-loader`
 
-    include the `FileDataLoader` struct. The `libextension_data_loader.a` static library is required, compile C++ `executorch` with `EXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON`.
+    Includes the `FileDataLoader` and `MmapDataLoader` structs. Without this feature the only available data loader is `BufferDataLoader`. The `libextension_data_loader.a` static library is required, compile C++ `executorch` with `EXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON`.
 
 - `module`
 
-    include the `Module` struct. The `libextension_module_static.a` static library is required, compile C++ `executorch` with `EXECUTORCH_BUILD_EXTENSION_MODULE=ON`.
+    Includes the `Module` struct. The `libextension_module_static.a` static library is required, compile C++ `executorch` with `EXECUTORCH_BUILD_EXTENSION_MODULE=ON`.
     Also includes the `std` feature.
 
 - `std`

executorch-sys/cpp/executorch_rs_ext/api_utils.cpp

@@ -1,5 +1,6 @@
 
 #include <cstddef>
+#include <vector>
 
 #include "executorch_rs_ext/api_utils.hpp"
 
@@ -163,12 +164,10 @@ namespace executorch_rs
         return list->get();
     }
 
-#if defined(EXECUTORCH_RS_DATA_LOADER)
     torch::executor::util::BufferDataLoader BufferDataLoader_new(const void *data, size_t size)
     {
         return torch::executor::util::BufferDataLoader(data, size);
     }
-#endif
 
 #if defined(EXECUTORCH_RS_MODULE)
     torch::executor::Module Module_new(torch::executor::ArrayRef<char> file_path, torch::executor::Module::MlockConfig mlock_config, torch::executor::EventTracer *event_tracer)

executorch-sys/cpp/executorch_rs_ext/api_utils.hpp

@@ -11,9 +11,7 @@
 #include "executorch/runtime/core/span.h"
 #include "executorch/runtime/core/exec_aten/exec_aten.h"
 
-#if defined(EXECUTORCH_RS_DATA_LOADER)
 #include "executorch/extension/data_loader/buffer_data_loader.h"
-#endif
 
 #if defined(EXECUTORCH_RS_MODULE)
 #include "executorch/extension/module/module.h"
@@ -90,9 +88,7 @@ namespace executorch_rs
     const exec_aten::ArrayRef<int64_t> BoxedEvalueList_i64_get(const torch::executor::BoxedEvalueList<int64_t> *list);
     const exec_aten::ArrayRef<exec_aten::Tensor> BoxedEvalueList_Tensor_get(const torch::executor::BoxedEvalueList<exec_aten::Tensor> *list);
 
-#if defined(EXECUTORCH_RS_DATA_LOADER)
     torch::executor::util::BufferDataLoader BufferDataLoader_new(const void *data, size_t size);
-#endif
 
 #if defined(EXECUTORCH_RS_MODULE)
     torch::executor::Module Module_new(torch::executor::ArrayRef<char> file_path, torch::executor::Module::MlockConfig mlock_config, torch::executor::EventTracer *event_tracer);

executorch-sys/src/lib.rs

@@ -6,7 +6,8 @@
 //!
 //!
 //! To build the library, you need to build the C++ library first.
-//! The C++ library allow for great flexibility with many flags, customizing which modules, kernels, and extensions are built.
+//! The C++ library allow for great flexibility with many flags, customizing which modules, kernels, and extensions are
+//! built.
 //! Multiple static libraries are built, and the Rust library links to them.
 //! In the following example we build the C++ library with the necessary flags to run example `hello_world_add`:
 //! ```bash
@@ -61,21 +62,27 @@
 //! - `libextension_data_loader.a`
 //! - `libextension_module_static.a`
 //!
-//! The static libraries of the kernels implementations are required only if your model uses them, and they should be **linked manually** by the binary that uses the `executorch` crate.
-//! For example, the `hello_world_add` example uses a model with a single addition operation, so it compile the C++ library with `DEXECUTORCH_SELECT_OPS_LIST=aten::add.out` and contain the following lines in its `build.rs`:
+//! The static libraries of the kernels implementations are required only if your model uses them, and they should be
+//! **linked manually** by the binary that uses the `executorch` crate.
+//! For example, the `hello_world_add` example uses a model with a single addition operation, so it compile the C++
+//! library with `DEXECUTORCH_SELECT_OPS_LIST=aten::add.out` and contain the following lines in its `build.rs`:
 //! ```rust
-//! println!("cargo::rustc-link-lib=static=portable_kernels");
+//! println!("cargo::rustc-link-lib=static:+whole-archive=portable_kernels");
 //! println!("cargo::rustc-link-lib=static:+whole-archive=portable_ops_lib");
 //!
 //! let libs_dir = std::env::var("EXECUTORCH_RS_EXECUTORCH_LIB_DIR").unwrap();
 //! println!("cargo::rustc-link-search={}/kernels/portable/", libs_dir);
 //! ```
-//! Note that the `portable_ops_lib` is linked with `+whole-archive` to ensure that all symbols are included in the binary.
+//! Note that the ops and kernels libs are linked with `+whole-archive` to ensure that all symbols are included in the
+//! binary.
 //!
 //! ## Cargo Features
 //! By default all features are disabled.
-//! - `data-loader`: include the `FileDataLoader` struct. The `libextension_data_loader.a` static library is required, compile C++ `executorch` with `EXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON`.
-//! - `module`: include the `Module` struct. The `libextension_module_static.a` static library is required, compile C++ `executorch` with `EXECUTORCH_BUILD_EXTENSION_MODULE=ON`.
+//! - `data-loader`: Includes the `FileDataLoader` and `MmapDataLoader` structs. Without this feature the only available
+//! data loader is `BufferDataLoader`. The `libextension_data_loader.a` static library is required, compile C++
+//! `executorch` with `EXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON`.
+//! - `module`: Includes the `Module` struct. The `libextension_module_static.a` static library is required, compile C++
+//! `executorch` with `EXECUTORCH_BUILD_EXTENSION_MODULE=ON`.
 
 #![cfg_attr(not(feature = "std"), no_std)]
 

scripts/requirements.txt

@@ -0,0 +1,9 @@
+cmake
+pyyaml
+setuptools>=63
+tomli
+wheel
+zstd
+torch==2.4.0
+# torchvision==0.19.0
+# torchaudio==2.4.0

scripts/setup_dev.py

@@ -19,6 +19,11 @@ def main():
         action="store_true",
         help="Remove the existing executorch directory before cloning",
     )
+    parser.add_argument(
+        "--skip-executorch-python",
+        action="store_true",
+        help="Remove the existing executorch directory before cloning",
+    )
     args = parser.parse_args()
 
     if args.clean:
@@ -28,23 +33,19 @@ def main():
     # TODO setup a venv here
 
     clone_executorch()
-
-    # subprocess.check_call(["./install_requirements.sh"], cwd=DEV_EXECUTORCH_DIR)
-    subprocess.check_call(
-        [
-            "pip",
-            "install",
+    if not args.skip_executorch_python:
+        subprocess.check_call(["./install_requirements.sh"], cwd=DEV_EXECUTORCH_DIR)
+    else:
+        deps = [
             "cmake",
             "pyyaml",
             "setuptools>=63",
             "tomli",
             "wheel",
             "zstd",
-            "torch==2.4.0",
         ]
-    )
-
-    build_executorch()
+        subprocess.check_call([get_pip(), "install", *deps])
+    build_executorch_with_dev_cfg()
 
 
 def clone_executorch():
@@ -70,23 +71,12 @@ def clone_executorch():
     subprocess.check_call(
         ["git", "submodule", "update", "--init"], cwd=DEV_EXECUTORCH_DIR
     )
-    # git submodule update --init backends/xnnpack/third-party/cpuinfo
-    # git submodule update --init backends/xnnpack/third-party/pthreadpool
-    # git submodule update --init third-party/prelude
-    # git submodule update --init third-party/gflags
-    # git submodule update --init third-party/googletest
-    # git submodule update --init third-party/flatbuffers
-    # git submodule update --init third-party/flatcc
-    # git submodule update --init backends/xnnpack/third-party/XNNPACK
-    # git submodule update --init backends/xnnpack/third-party/FXdiv
-    # git submodule update --init third-party/pytorch
-
-
-def build_executorch():
+
+
+def build_executorch_with_dev_cfg():
     cmake_out_dir = DEV_EXECUTORCH_DIR / "cmake-out"
-    if cmake_out_dir.exists():
-        shutil.rmtree(cmake_out_dir)
-    cmake_out_dir.mkdir()
+    if not cmake_out_dir.exists():
+        cmake_out_dir.mkdir()
     subprocess.check_call(
         [
             "cmake",
@@ -110,5 +100,20 @@ def build_executorch():
     )
 
 
+def get_pip():
+    try:
+        subprocess.run(['pip3', '--version'], check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+        return 'pip3'
+    except (subprocess.CalledProcessError, FileNotFoundError):
+        pass
+
+    try:
+        subprocess.run(['pip', '--version'], check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+        return 'pip'
+    except (subprocess.CalledProcessError, FileNotFoundError):
+        pass
+
+    raise RuntimeError("Neither 'pip3' nor 'pip' is installed on this system.")
+
 if __name__ == "__main__":
     main()

src/data_loader.rs

@@ -4,27 +4,53 @@
 //! To include the data loader functionality, enable the `data-loader` feature.
 
 use std::cell::UnsafeCell;
+use std::marker::PhantomData;
 
-use crate::et_c;
+use crate::{et_c, et_rs_c};
 
 /// Loads from a data source.
 ///
 /// This struct is like a base class for data loaders. All other data loaders implement `AsRef<DataLoader>` and other
 /// structs, such as `Program`, take a reference to `DataLoader` instead of the concrete data loader type.
 pub struct DataLoader(pub(crate) UnsafeCell<et_c::DataLoader>);
 
+/// A DataLoader that wraps a pre-allocated buffer. The FreeableBuffers
+/// that it returns do not actually free any data.
+///
+/// This can be used to wrap data that is directly embedded into the firmware
+/// image, or to wrap data that was allocated elsewhere.
+#[allow(dead_code)]
+pub struct BufferDataLoader<'a>(
+    UnsafeCell<et_c::util::BufferDataLoader>,
+    PhantomData<&'a ()>,
+);
+impl<'a> BufferDataLoader<'a> {
+    /// Creates a new BufferDataLoader that wraps the given data.
+    pub fn new(data: &'a [u8]) -> Self {
+        let loader =
+            unsafe { et_rs_c::BufferDataLoader_new(data.as_ptr() as *const _, data.len()) };
+        Self(UnsafeCell::new(loader), PhantomData)
+    }
+}
+impl AsRef<DataLoader> for BufferDataLoader<'_> {
+    fn as_ref(&self) -> &DataLoader {
+        // SAFETY: BufferDataLoader has a single field of (UnsafeCell of) et_c::util::BufferDataLoader, which is a
+        // subclass of et_c::DataLoader, and DataLoaders has a single field of (UnsafeCell of) et_c::DataLoader.
+        unsafe { std::mem::transmute::<&BufferDataLoader, &DataLoader>(self) }
+    }
+}
+
 #[cfg(feature = "data-loader")]
-pub use file_data_loader::{BufferDataLoader, FileDataLoader, MlockConfig, MmapDataLoader};
+pub use file_data_loader::{FileDataLoader, MlockConfig, MmapDataLoader};
 
 #[cfg(feature = "data-loader")]
 mod file_data_loader {
     use std::cell::UnsafeCell;
     use std::ffi::CStr;
-    use std::marker::PhantomData;
 
     use crate::error::Result;
+    use crate::et_c;
     use crate::util::IntoRust;
-    use crate::{et_c, et_rs_c};
 
     use super::DataLoader;
 
@@ -182,32 +208,6 @@ mod file_data_loader {
         }
     }
 
-    /// A DataLoader that wraps a pre-allocated buffer. The FreeableBuffers
-    /// that it returns do not actually free any data.
-    ///
-    /// This can be used to wrap data that is directly embedded into the firmware
-    /// image, or to wrap data that was allocated elsewhere.
-    #[allow(dead_code)]
-    pub struct BufferDataLoader<'a>(
-        UnsafeCell<et_c::util::BufferDataLoader>,
-        PhantomData<&'a ()>,
-    );
-    impl<'a> BufferDataLoader<'a> {
-        /// Creates a new BufferDataLoader that wraps the given data.
-        pub fn new(data: &'a [u8]) -> Self {
-            let loader =
-                unsafe { et_rs_c::BufferDataLoader_new(data.as_ptr() as *const _, data.len()) };
-            Self(UnsafeCell::new(loader), PhantomData)
-        }
-    }
-    impl AsRef<DataLoader> for BufferDataLoader<'_> {
-        fn as_ref(&self) -> &DataLoader {
-            // SAFETY: BufferDataLoader has a single field of (UnsafeCell of) et_c::util::BufferDataLoader, which is a
-            // subclass of et_c::DataLoader, and DataLoaders has a single field of (UnsafeCell of) et_c::DataLoader.
-            unsafe { std::mem::transmute::<&BufferDataLoader, &DataLoader>(self) }
-        }
-    }
-
     /// Describes how and whether to lock loaded pages with `mlock()`.
     ///
     /// Using `mlock()` typically loads all of the pages immediately, and will