onednn-src 0.1.13

Source of oneAPI Deep Neural Network Library (oneDNN)
Documentation
/*******************************************************************************
* Copyright 2019 Intel Corporation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
*     http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*******************************************************************************/

#ifndef XPU_SYCL_USM_MEMORY_STORAGE_HPP
#define XPU_SYCL_USM_MEMORY_STORAGE_HPP

#include "oneapi/dnnl/dnnl_config.h"

#include "common/engine.hpp"
#include "common/memory_storage.hpp"
#include "common/utils.hpp"

#include "xpu/sycl/engine_impl.hpp"
#include "xpu/sycl/memory_storage_base.hpp"

#include <memory>

namespace dnnl {
namespace impl {
namespace xpu {
namespace sycl {

class usm_memory_storage_t : public memory_storage_base_t {
public:
    using memory_storage_base_t::memory_storage_base_t;

    usm_memory_storage_t(engine_t *engine, ::sycl::usm::alloc usm_kind)
        : memory_storage_base_t(engine), usm_kind_(usm_kind) {}
    ~usm_memory_storage_t() override = default;

    uint8_t *usm_ptr() const { return static_cast<uint8_t *>(usm_ptr_.get()); }

    memory_kind_t memory_kind() const override { return memory_kind::usm; }

    status_t get_data_handle(void **handle) const override {
        *handle = usm_ptr_.get();
        return status::success;
    }

    status_t set_data_handle(void *handle) override {
        const auto *sycl_engine_impl
                = utils::downcast<const xpu::sycl::engine_impl_t *>(
                        engine()->impl());
        auto &sycl_ctx = sycl_engine_impl->context();

        usm_ptr_ = decltype(usm_ptr_)(handle, [](void *) {});
        usm_kind_ = ::sycl::get_pointer_type(handle, sycl_ctx);

        return status::success;
    }

    status_t map_data(
            void **mapped_ptr, stream_t *stream, size_t size) const override;
    status_t unmap_data(void *mapped_ptr, stream_t *stream) const override;

    bool is_host_accessible() const override {
        if (engine()->kind() == engine_kind::cpu) return true; // optimism

        return utils::one_of(usm_kind_, ::sycl::usm::alloc::host,
                ::sycl::usm::alloc::shared, ::sycl::usm::alloc::unknown);
    }

    std::unique_ptr<memory_storage_t> get_sub_storage(
            size_t offset, size_t size) const override {
        void *sub_ptr = usm_ptr_
                ? reinterpret_cast<uint8_t *>(usm_ptr_.get()) + offset
                : nullptr;
        auto storage = utils::make_unique<usm_memory_storage_t>(engine());
        if (!storage) return nullptr;
        storage->init(memory_flags_t::use_runtime_ptr, size, sub_ptr);
        return storage;
    }

    std::unique_ptr<memory_storage_t> clone() const override {
        auto storage = utils::make_unique<usm_memory_storage_t>(engine());
        if (!storage) return nullptr;

        status_t status
                = storage->init(memory_flags_t::use_runtime_ptr, 0, nullptr);
        if (status != status::success) return nullptr;

        storage->usm_ptr_ = decltype(usm_ptr_)(usm_ptr_.get(), [](void *) {});
        storage->usm_kind_ = usm_kind_;
        storage->set_offset(offset());

        return storage;
    }

    std::unique_ptr<memory_storage_t> clone_ptr_off(
            size_t offset) const override {
        return get_sub_storage(offset, 0);
    }

    in_memory_arg_t get_in_memory_arg(
            stream_t *stream, ::sycl::handler &cgh) const override;
    out_memory_arg_t get_out_memory_arg(
            stream_t *stream, ::sycl::handler &cgh) const override;
    inout_memory_arg_t get_inout_memory_arg(
            stream_t *stream, ::sycl::handler &cgh) const override;

protected:
    status_t init_allocate(size_t size) override {
        const auto *sycl_engine_impl
                = utils::downcast<const xpu::sycl::engine_impl_t *>(
                        engine()->impl());
        auto &sycl_dev = sycl_engine_impl->device();
        auto &sycl_ctx = sycl_engine_impl->context();
        using ::sycl::usm::alloc;

        if (usm_kind_ == alloc::unknown) usm_kind_ = alloc::device;

        void *usm_ptr_alloc = nullptr;

        switch (usm_kind_) {
            case alloc::host:
                usm_ptr_alloc = ::sycl::malloc_host(size, sycl_ctx);
                break;
            case alloc::device:
                usm_ptr_alloc = ::sycl::malloc_device(size, sycl_dev, sycl_ctx);
                break;
            case alloc::shared:
                usm_ptr_alloc = ::sycl::malloc_shared(size, sycl_dev, sycl_ctx);
                break;
            default: break;
        }
        if (!usm_ptr_alloc) return status::out_of_memory;

        usm_ptr_ = decltype(usm_ptr_)(
                usm_ptr_alloc, [&](void *ptr) { ::sycl::free(ptr, sycl_ctx); });
        return status::success;
    }

private:
    std::unique_ptr<void, std::function<void(void *)>> usm_ptr_;
    ::sycl::usm::alloc usm_kind_ = ::sycl::usm::alloc::unknown;
};

} // namespace sycl
} // namespace xpu
} // namespace impl
} // namespace dnnl

#endif // XPU_SYCL_USM_MEMORY_STORAGE_HPP