mmdeploy/csrc/core/device_impl.h

187 lines
4.6 KiB
C
Raw Normal View History

Merge sdk (#251) * check in cmake * move backend_ops to csrc/backend_ops * check in preprocess, model, some codebase and their c-apis * check in CMakeLists.txt * check in parts of test_csrc * commit everything else * add readme * update core's BUILD_INTERFACE directory * skip codespell on third_party * update trt_net and ort_net's CMakeLists * ignore clion's build directory * check in pybind11 * add onnx.proto. Remove MMDeploy's dependency on ncnn's source code * export MMDeployTargets only when MMDEPLOY_BUILD_SDK is ON * remove useless message * target include directory is wrong * change target name from mmdeploy_ppl_net to mmdeploy_pplnn_net * skip install directory * update project's cmake * remove useless code * set CMAKE_BUILD_TYPE to Release by force if it isn't set by user * update custom ops CMakeLists * pass object target's source lists * fix lint end-of-file * fix lint: trailing whitespace * fix codespell hook * remove bicubic_interpolate to csrc/backend_ops/ * set MMDEPLOY_BUILD_SDK OFF * change custom ops build command * add spdlog installation command * update docs on how to checkout pybind11 * move bicubic_interpolate to backend_ops/tensorrt directory * remove useless code * correct cmake * fix typo * fix typo * fix install directory * correct sdk's readme * set cub dir when cuda version < 11.0 * change directory where clang-format will apply to * fix build command * add .clang-format * change clang-format style from google to file * reformat csrc/backend_ops * format sdk's code * turn off clang-format for some files * add -Xcompiler=-fno-gnu-unique * fix trt topk initialize * check in config for sdk demo * update cmake script and csrc's readme * correct config's path * add cuda include directory, otherwise compile failed in case of tensorrt8.2 * clang-format onnx2ncnn.cpp Co-authored-by: zhangli <lzhang329@gmail.com> Co-authored-by: grimoire <yaoqian@sensetime.com>
2021-12-07 10:57:55 +08:00
// Copyright (c) OpenMMLab. All rights reserved.
#ifndef MMDEPLOY_SRC_CORE_DEVICE_IMPL_H_
#define MMDEPLOY_SRC_CORE_DEVICE_IMPL_H_
#include "device.h"
namespace mmdeploy {
using std::shared_ptr;
using PlatformImplPtr = shared_ptr<PlatformImpl>;
using AllocatorImplPtr = shared_ptr<AllocatorImpl>;
using BufferImplPtr = shared_ptr<BufferImpl>;
using StreamImplPtr = shared_ptr<StreamImpl>;
using EventImplPtr = shared_ptr<EventImpl>;
class PlatformImpl {
public:
PlatformImpl() : platform_id_(-1) {}
virtual ~PlatformImpl() = default;
virtual const char* GetPlatformName() const noexcept = 0;
virtual int GetPlatformId() const noexcept { return platform_id_; }
virtual void SetPlatformId(int id) { platform_id_ = id; }
virtual Result<void> SetDevice(Device device) { return success(); };
virtual shared_ptr<BufferImpl> CreateBuffer(Device device) = 0;
virtual shared_ptr<StreamImpl> CreateStream(Device device) = 0;
virtual shared_ptr<EventImpl> CreateEvent(Device device) = 0;
virtual Result<void> Copy(const void* host_ptr, Buffer dst, size_t size, size_t dst_offset,
Stream stream) = 0;
virtual Result<void> Copy(Buffer src, void* host_ptr, size_t size, size_t src_offset,
Stream stream) = 0;
virtual Result<void> Copy(Buffer src, Buffer dst, size_t size, size_t src_offset,
size_t dst_offset, Stream stream) = 0;
virtual Result<Stream> GetDefaultStream(int32_t device_id) = 0;
protected:
int platform_id_;
};
class AllocatorImpl {
public:
struct Block {
explicit Block(void* _handle = nullptr, size_t _size = 0) : handle(_handle), size(_size) {}
void* handle;
size_t size;
};
virtual ~AllocatorImpl() = default;
virtual Block Allocate(size_t size) noexcept = 0;
virtual void Deallocate(Block& block) noexcept = 0;
virtual bool Owns(const Block& block) const noexcept = 0;
virtual const char* Name() const noexcept { return ""; }
// virtual Device device() const noexcept = 0;
};
// create, destroy, sub, MakeAvailableOnDevice, FromHost, fill, copy, map, unmap
class BufferImpl {
public:
explicit BufferImpl(Device device) : device_(device) {}
virtual ~BufferImpl() = default;
virtual Result<void> Init(size_t size, Allocator allocator, size_t alignment, uint64_t flags) = 0;
virtual Result<void> Init(size_t size, std::shared_ptr<void> native, uint64_t flags) = 0;
virtual Result<shared_ptr<BufferImpl>> SubBuffer(size_t offset, size_t size, uint64_t flags) = 0;
virtual size_t GetSize(ErrorCode* ec) = 0;
virtual Allocator GetAllocator() const = 0;
virtual void* GetNative(ErrorCode* ec) = 0;
Device GetDevice() const noexcept { return device_; }
protected:
Device device_;
};
class StreamImpl {
public:
explicit StreamImpl(Device device) : device_(device) {}
virtual ~StreamImpl() = default;
virtual Result<void> Init(uint64_t flags) = 0;
virtual Result<void> Init(std::shared_ptr<void> native, uint64_t flags) = 0;
virtual Result<void> Query() = 0;
virtual Result<void> Wait() = 0;
virtual Result<void> Submit(Kernel& kernel) = 0;
virtual Result<void> DependsOn(Event& event) = 0;
virtual void* GetNative(ErrorCode* ec) = 0;
Device GetDevice() const noexcept { return device_; }
protected:
Device device_;
};
class EventImpl {
public:
explicit EventImpl(Device device) : device_(device) {}
virtual ~EventImpl() = default;
virtual Result<void> Init(uint64_t flags) = 0;
virtual Result<void> Init(std::shared_ptr<void> native, uint64_t flags) = 0;
virtual Result<void> Query() = 0;
virtual Result<void> Record(Stream& st) = 0;
virtual Result<void> Wait() = 0;
virtual void* GetNative(ErrorCode* ec) = 0;
Device GetDevice() const noexcept { return device_; }
protected:
Device device_;
};
class KernelWrapper {
public:
virtual ~KernelWrapper() = default;
virtual int Invoke(const std::vector<void*>& args) = 0;
};
class KernelImpl {
public:
explicit KernelImpl(Device device) : device_(device) {}
virtual ~KernelImpl() = default;
Device GetDevice() const noexcept { return device_; }
virtual void* GetNative(ErrorCode* ec) = 0;
protected:
Device device_;
};
struct Access {
template <typename T, typename Obj>
static T& get(const Obj& obj) {
return static_cast<T&>(*obj.impl_);
}
template <typename Obj>
static auto& get_impl(const Obj& obj) {
return obj.impl_;
}
template <typename T, typename... Args>
static T create(Args&&... args) {
return T(std::forward<Args>(args)...);
}
};
inline PlatformImpl* GetPlatformImpl(const Device& device) {
return gPlatformRegistry().GetPlatformImpl(device);
}
} // namespace mmdeploy
#endif