Bump Cpp lib to 1.4.0

.github/workflows/ci_requirements.txt

@@ -1,4 +1,4 @@
-# Copy of executorch/requirements-dev.txt as of v1.3.1
+# Copy of executorch/requirements-dev.txt as of v1.4.0
 
 cmake>=3.29, <4.0.0  # For building binary targets in the wheel.
 packaging>=24.2 # Lower bound required by setuptools
@@ -10,4 +10,4 @@ zstd  # Imported by resolve_buck.py.
 certifi  # Imported by resolve_buck.py.
 lintrunner==0.12.7
 lintrunner-adapters==0.13.0
-torch==2.12.0
+torch==2.13.0

README.md

@@ -63,7 +63,7 @@ See `example/hello_world` for a complete example.
 ## Build
 To use the library you must compile the C++ executorch library yourself, as there are many configurations that
 determines which modules, backends, and operations are supported. See the `executorch-sys` crate for more info.
-Currently the supported Cpp executorch version is `1.3.1`.
+Currently the supported Cpp executorch version is `1.4.0`.
 
 The table below shows which version of the Rust crate is compatible with which C++ ExecuTorch version:
 

etc/download_headers.py

@@ -63,8 +63,6 @@ def main():
             "extension/data_loader/mman.h",
             "extension/data_loader/mman_windows.h",
             "extension/module/bundled_module.h",  # TODO
-            "runtime/core/device_allocator.h",  # TODO
-            "runtime/core/device_memory_buffer.h",  # TODO
             "extension/flat_tensor/serialize/serialize.h",
             "**/test/**",
             "**/testing_util/**",
@@ -74,9 +72,7 @@ def main():
         for include in includes:
             files.update(glob.glob(include, root_dir=tmp_headers_dir, recursive=True))
         for exclude in excludes:
-            files.difference_update(
-                glob.glob(exclude, root_dir=tmp_headers_dir, recursive=True)
-            )
+            files.difference_update(glob.glob(exclude, root_dir=tmp_headers_dir, recursive=True))
         for f in files:
             dst = HEADERS_DIR / f
             if not dst.parent.exists():

etc/setup_dev_env.py

@@ -7,9 +7,7 @@
 import warnings
 from pathlib import Path
 
-DEV_EXECUTORCH_DIR = (
-    Path(__file__).parent.parent.resolve() / "etc" / ".dev-env" / "executorch"
-)
+DEV_EXECUTORCH_DIR = Path(__file__).parent.parent.resolve() / "etc" / ".dev-env" / "executorch"
 
 
 def main():
@@ -49,16 +47,14 @@ def main():
                 "install",
                 "-r",
                 DEV_EXECUTORCH_DIR / "requirements-dev.txt",
-                "torch==2.12.0",
+                "torch==2.13.0",
                 "--extra-index-url",
                 "https://download.pytorch.org/whl/test/cpu",
             ]
         )
     build_executorch_with_dev_cfg()
 
-    subprocess.check_call(
-        [sys.executable, "-m", "pip", "install", "huggingface_hub[cli]"]
-    )
+    subprocess.check_call([sys.executable, "-m", "pip", "install", "huggingface_hub[cli]"])
 
 
 def clone_executorch():
@@ -71,7 +67,7 @@ def clone_executorch():
                 "--depth",
                 "1",
                 "--branch",
-                "v1.3.1",
+                "v1.4.0",
                 "https://github.com/pytorch/executorch.git",
                 ".",
             ],
@@ -94,12 +90,8 @@ def clone_executorch():
                 cwd=DEV_EXECUTORCH_DIR / "backends" / "apple" / "coreml" / "scripts",
             )
 
-    subprocess.check_call(
-        ["git", "submodule", "update", "--init", "--recursive"], cwd=DEV_EXECUTORCH_DIR
-    )
-    subprocess.check_call(
-        ["git", "submodule", "sync", "--recursive"], cwd=DEV_EXECUTORCH_DIR
-    )
+    subprocess.check_call(["git", "submodule", "update", "--init", "--recursive"], cwd=DEV_EXECUTORCH_DIR)
+    subprocess.check_call(["git", "submodule", "sync", "--recursive"], cwd=DEV_EXECUTORCH_DIR)
 
 
 def build_executorch_with_dev_cfg():

examples/llama3/README.md

@@ -72,4 +72,4 @@ To run the example, follow these steps (note that some steps should run from the
 The llama model can be exported with many options, such as quantization, different data types (f32, bf16), different backends, kv caching, etc.
 This example use a specific set of options, as specified above.
 Different options require different export commands and modifications to the code and build script which you can play around with.
-See the [llama README](https://github.com/pytorch/executorch/blob/v1.3.1/examples/models/llama/README.md) at the Cpp executorch repository for more details.
+See the [llama README](https://github.com/pytorch/executorch/blob/v1.4.0/examples/models/llama/README.md) at the Cpp executorch repository for more details.

executorch-sys/README.md

@@ -4,14 +4,14 @@ For a general description of the project, see the the `executorch` crate.
 
 ## Build
 To build the library, you need to build the C++ library yourself first.
-Currently the supported Cpp executorch version is `1.3.1`.
+Currently the supported Cpp executorch version is `1.4.0`.
 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`:
 ```bash
 # Clone the C++ library
 cd ${EXECUTORCH_CPP_DIR}
-git clone --depth 1 --branch v1.3.1 https://github.com/pytorch/executorch.git .
+git clone --depth 1 --branch v1.4.0 https://github.com/pytorch/executorch.git .
 git submodule sync --recursive
 git submodule update --init --recursive
 

executorch-sys/build.rs

@@ -1,6 +1,6 @@
 use std::path::{Path, PathBuf};
 
-// const EXECUTORCH_VERSION: &str = "1.3.1";
+// const EXECUTORCH_VERSION: &str = "1.4.0";
 
 fn main() {
     // TODO: verify on runtime we use the correct version of executorch

executorch-sys/cpp/executorch_rs/c_bridge.cpp

@@ -67,6 +67,10 @@ namespace
 
     static_assert(is_equal_layout<struct ET_Device, executorch::runtime::etensor::Device>());
     static_assert(std::is_trivially_move_constructible_v<executorch::runtime::etensor::Device>);
+    static_assert(is_equal_layout<enum ET_DeviceType, executorch::runtime::etensor::DeviceType>());
+    static_assert(std::is_trivially_move_constructible_v<executorch::runtime::etensor::DeviceType>);
+    static_assert(is_equal_layout<int8_t, executorch::runtime::etensor::DeviceIndex>());
+    static_assert(std::is_trivially_move_constructible_v<executorch::runtime::etensor::DeviceIndex>);
 
     static_assert(is_equal_layout<struct ET_BackendOption, executorch::runtime::BackendOption>());
     static_assert(std::is_trivially_move_constructible_v<executorch::runtime::BackendOption>);
@@ -129,6 +133,7 @@ namespace
     //
     // static_assert(std::is_trivially_move_constructible_v<executorch::runtime::MemoryAllocator>);
 
+    static_assert(is_equal_layout<struct ET_SpanDevice, executorch::runtime::Span<const executorch::runtime::etensor::Device>>());
     static_assert(is_equal_layout<struct ET_HierarchicalAllocator, executorch::runtime::HierarchicalAllocator>());
     static_assert(std::is_trivially_move_constructible_v<executorch::runtime::HierarchicalAllocator>);
 
@@ -200,6 +205,16 @@ void *executorch_MemoryAllocator_allocate(struct ET_MemoryAllocator *self, size_
     auto self_ = checked_reinterpret_cast<executorch::runtime::MemoryAllocator>(self);
     return self_->allocate(size, alignment);
 }
+size_t executorch_MemoryAllocator_used_size(const struct ET_MemoryAllocator *self)
+{
+    auto self_ = checked_reinterpret_cast<const executorch::runtime::MemoryAllocator>(self);
+    return self_->used_size();
+}
+size_t executorch_MemoryAllocator_free_size(const struct ET_MemoryAllocator *self)
+{
+    auto self_ = checked_reinterpret_cast<const executorch::runtime::MemoryAllocator>(self);
+    return self_->free_size();
+}
 struct ET_HierarchicalAllocator executorch_HierarchicalAllocator_new(struct ET_SpanSpanU8 buffers)
 {
     auto buffers_ = *checked_reinterpret_cast<executorch::runtime::Span<executorch::runtime::Span<uint8_t>>>(&buffers);
@@ -210,6 +225,28 @@ struct ET_HierarchicalAllocator executorch_HierarchicalAllocator_new(struct ET_S
     new (self_) executorch::runtime::HierarchicalAllocator(buffers_);
     return self;
 }
+struct ET_HierarchicalAllocator executorch_HierarchicalAllocator_new_with_devices(struct ET_SpanSpanU8 buffers, struct ET_SpanDevice planned_buffer_devices)
+{
+    auto buffers_ = *checked_reinterpret_cast<executorch::runtime::Span<executorch::runtime::Span<uint8_t>>>(&buffers);
+    ET_CHECK((void *)buffers_.begin() == (void *)buffers.data);
+    ET_CHECK(buffers_.size() == buffers.len);
+    auto devices_ = *checked_reinterpret_cast<executorch::runtime::Span<const executorch::runtime::etensor::Device>>(&planned_buffer_devices);
+    ET_CHECK((const void *)devices_.begin() == (const void *)planned_buffer_devices.data);
+    ET_CHECK(devices_.size() == planned_buffer_devices.len);
+    struct ET_HierarchicalAllocator self;
+    auto self_ = checked_reinterpret_cast<executorch::runtime::HierarchicalAllocator>(&self);
+    new (self_) executorch::runtime::HierarchicalAllocator(buffers_, devices_);
+    return self;
+}
+struct ET_SpanDevice executorch_HierarchicalAllocator_planned_buffer_devices(const struct ET_HierarchicalAllocator *self)
+{
+    auto self_ = checked_reinterpret_cast<const executorch::runtime::HierarchicalAllocator>(self);
+    auto devices = self_->planned_buffer_devices();
+    return ET_SpanDevice{
+        .data = reinterpret_cast<const struct ET_Device *>(devices.data()),
+        .len = devices.size(),
+    };
+}
 void executorch_HierarchicalAllocator_destructor(struct ET_HierarchicalAllocator *self)
 {
     auto self_ = checked_reinterpret_cast<executorch::runtime::HierarchicalAllocator>(self);
@@ -229,6 +266,20 @@ struct ET_MemoryManager executorch_MemoryManager_new(
     new (self_) executorch::runtime::MemoryManager(method_allocator_, planned_memory_, temp_allocator_);
     return self;
 }
+struct ET_SpanDevice executorch_MemoryManager_planned_buffer_devices(const struct ET_MemoryManager *self)
+{
+    auto self_ = checked_reinterpret_cast<const executorch::runtime::MemoryManager>(self);
+    auto devices = self_->planned_buffer_devices();
+    return ET_SpanDevice{
+        .data = reinterpret_cast<const struct ET_Device *>(devices.data()),
+        .len = devices.size(),
+    };
+}
+bool executorch_MemoryManager_has_device_memory(const struct ET_MemoryManager *self)
+{
+    auto self_ = checked_reinterpret_cast<const executorch::runtime::MemoryManager>(self);
+    return self_->has_device_memory();
+}
 
 // Loaders
 static executorch::runtime::DataLoader *cast_data_loader_mut(struct ET_DataLoaderRefMut loader)
@@ -448,7 +499,7 @@ enum ET_ScalarType executorch_Tensor_scalar_type(struct ET_TensorRef self)
 struct ET_Device executorch_Tensor_device(struct ET_TensorRef self)
 {
     auto self_ = cast_tensor(self);
-    auto d = self_->unsafeGetTensorImpl()->device();
+    auto d = self_->device();
     return ET_Device{static_cast<ET_DeviceType>(d.type()), static_cast<int8_t>(d.index())};
 }
 size_t executorch_Tensor_element_size(struct ET_TensorRef self)

executorch-sys/cpp/executorch_rs/c_bridge.h

@@ -505,7 +505,7 @@ extern "C"
     };
     struct ET_HierarchicalAllocator
     {
-        size_t _blob[34];
+        size_t _blob[36];
     };
     struct ET_MemoryManager
     {
@@ -631,6 +631,11 @@ extern "C"
         uint8_t *data;
         size_t len;
     };
+    struct ET_SpanDevice
+    {
+        const struct ET_Device *data;
+        size_t len;
+    };
     struct ET_SpanSpanU8
     {
         struct ET_SpanU8 *data;
@@ -712,12 +717,19 @@ extern "C"
 
     struct ET_MemoryAllocator executorch_MemoryAllocator_new(uint32_t size, uint8_t *base_address);
     void *executorch_MemoryAllocator_allocate(struct ET_MemoryAllocator *self, size_t size, size_t alignment);
+    size_t executorch_MemoryAllocator_used_size(const struct ET_MemoryAllocator *self);
+    size_t executorch_MemoryAllocator_free_size(const struct ET_MemoryAllocator *self);
     struct ET_HierarchicalAllocator executorch_HierarchicalAllocator_new(struct ET_SpanSpanU8 buffers);
+    struct ET_HierarchicalAllocator executorch_HierarchicalAllocator_new_with_devices(struct ET_SpanSpanU8 buffers, struct ET_SpanDevice planned_buffer_devices);
+    struct ET_SpanDevice executorch_HierarchicalAllocator_planned_buffer_devices(
+        const struct ET_HierarchicalAllocator *self);
     void executorch_HierarchicalAllocator_destructor(struct ET_HierarchicalAllocator *self);
     struct ET_MemoryManager executorch_MemoryManager_new(
         struct ET_MemoryAllocator *method_allocator,
         struct ET_HierarchicalAllocator *planned_memory,
         struct ET_MemoryAllocator *temp_allocator);
+    struct ET_SpanDevice executorch_MemoryManager_planned_buffer_devices(const struct ET_MemoryManager *self);
+    bool executorch_MemoryManager_has_device_memory(const struct ET_MemoryManager *self);
 
     // Loaders
     struct ET_BufferDataLoader executorch_BufferDataLoader_new(const void *data, size_t size);

executorch-sys/cpp/executorch_rs/cxx_bridge.cpp

@@ -70,7 +70,8 @@ namespace executorch_rs
         std::unique_ptr<std::vector<int32_t>> strides,
         ET_ScalarType scalar_type,
         ET_TensorShapeDynamism dynamism,
-        rust::Box<executorch_rs::cxx_util::RustAny> allocation)
+        rust::Box<executorch_rs::cxx_util::RustAny> allocation,
+        struct ET_Device device)
     {
         // std::function must be copyable, so we need to wrap the allocation in a shared_ptr
         std::shared_ptr<rust::Box<executorch_rs::cxx_util::RustAny>> allocation_ptr =
@@ -83,7 +84,10 @@ namespace executorch_rs
             std::move(*strides),
             static_cast<executorch::aten::ScalarType>(scalar_type),
             static_cast<executorch::aten::TensorShapeDynamism>(dynamism),
-            [allocation_ptr = allocation_ptr](void *) mutable {});
+            [allocation_ptr = allocation_ptr](void *) mutable {},
+            executorch::aten::Device(
+                static_cast<executorch::aten::DeviceType>(device.type),
+                device.index));
     }
 
     std::shared_ptr<executorch::aten::Tensor> TensorPtr_clone(
@@ -93,6 +97,17 @@ namespace executorch_rs
         return executorch::extension::clone_tensor_ptr(
             tensor, static_cast<executorch::aten::ScalarType>(scalar_type));
     }
+
+    std::shared_ptr<executorch::aten::Tensor> TensorPtr_clone_to(
+        std::shared_ptr<executorch::aten::Tensor> tensor,
+        struct ET_Device device)
+    {
+        return executorch::extension::clone_tensor_ptr_to(
+            tensor,
+            executorch::aten::Device(
+                static_cast<executorch::aten::DeviceType>(device.type),
+                device.index));
+    }
 #endif
 
 #if defined(EXECUTORCH_RS_MODULE)

executorch-sys/cpp/executorch_rs/cxx_bridge.hpp

@@ -36,10 +36,14 @@ namespace executorch_rs
         std::unique_ptr<std::vector<int32_t>> strides,
         ET_ScalarType scalar_type,
         ET_TensorShapeDynamism dynamism,
-        rust::Box<executorch_rs::cxx_util::RustAny> allocation);
+        rust::Box<executorch_rs::cxx_util::RustAny> allocation,
+        struct ET_Device device);
     std::shared_ptr<executorch::aten::Tensor> TensorPtr_clone(
         const executorch::aten::Tensor &tensor,
         ET_ScalarType scalar_type);
+    std::shared_ptr<executorch::aten::Tensor> TensorPtr_clone_to(
+        std::shared_ptr<executorch::aten::Tensor> tensor,
+        struct ET_Device device);
 #endif
 
 #if defined(EXECUTORCH_RS_MODULE)

executorch-sys/src/c_bridge.rs

@@ -60,6 +60,7 @@ impl_ref_clone_copy!(ET_ArrayRefU8);
 impl_ref_clone_copy!(ET_ArrayRefUsizeType);
 
 // Span
+impl_ref_clone_copy!(ET_SpanDevice);
 impl_ref_clone_copy!(ET_SpanI64);
 impl_ref_clone_copy!(ET_SpanOptionalTensor);
 impl_ref_clone_copy!(ET_SpanSpanU8);

executorch-sys/src/cxx_bridge/core.rs

@@ -93,3 +93,8 @@ unsafe impl ExternType for crate::ET_MemoryAllocator {
     type Id = type_id!("ET_MemoryAllocator");
     type Kind = cxx::kind::Opaque;
 }
+
+unsafe impl ExternType for crate::ET_Device {
+    type Id = type_id!("ET_Device");
+    type Kind = cxx::kind::Trivial;
+}

executorch-sys/src/cxx_bridge/mod.rs

@@ -17,4 +17,4 @@ pub use module::ffi::{
 #[cfg(feature = "tensor-ptr")]
 pub(crate) mod tensor_ptr;
 #[cfg(feature = "tensor-ptr")]
-pub use tensor_ptr::ffi::{Tensor, TensorPtr_clone, TensorPtr_new};
+pub use tensor_ptr::ffi::{Tensor, TensorPtr_clone, TensorPtr_clone_to, TensorPtr_new};