You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

344 lines
14 KiB

/*
* Copyright (C) 2019 The Android Open Source Project
*
* 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 ANDROID_FRAMEWORKS_ML_NN_COMMON_EXECUTION_BURST_SERVER_H
#define ANDROID_FRAMEWORKS_ML_NN_COMMON_EXECUTION_BURST_SERVER_H
#include <android-base/macros.h>
#include <android/hardware/neuralnetworks/1.0/types.h>
#include <android/hardware/neuralnetworks/1.1/types.h>
#include <android/hardware/neuralnetworks/1.2/IBurstCallback.h>
#include <android/hardware/neuralnetworks/1.2/IPreparedModel.h>
#include <android/hardware/neuralnetworks/1.2/types.h>
#include <fmq/MessageQueue.h>
#include <hidl/MQDescriptor.h>
#include <atomic>
#include <chrono>
#include <memory>
#include <optional>
#include <thread>
#include <tuple>
#include <vector>
namespace android::nn {
using FmqRequestDescriptor =
hardware::MQDescriptorSync<hardware::neuralnetworks::V1_2::FmqRequestDatum>;
using FmqResultDescriptor =
hardware::MQDescriptorSync<hardware::neuralnetworks::V1_2::FmqResultDatum>;
/**
* Function to serialize results.
*
* Prefer calling ResultChannelSender::send.
*
* @param errorStatus Status of the execution.
* @param outputShapes Dynamic shapes of the output tensors.
* @param timing Timing information of the execution.
* @return Serialized FMQ result data.
*/
std::vector<hardware::neuralnetworks::V1_2::FmqResultDatum> serialize(
hardware::neuralnetworks::V1_0::ErrorStatus errorStatus,
const std::vector<hardware::neuralnetworks::V1_2::OutputShape>& outputShapes,
hardware::neuralnetworks::V1_2::Timing timing);
/**
* Deserialize the FMQ request data.
*
* The three resulting fields are the Request object (where Request::pools is
* empty), slot identifiers (which are stand-ins for Request::pools), and
* whether timing information must be collected for the run.
*
* @param data Serialized FMQ request data.
* @return Request object if successfully deserialized, std::nullopt otherwise.
*/
std::optional<std::tuple<hardware::neuralnetworks::V1_0::Request, std::vector<int32_t>,
hardware::neuralnetworks::V1_2::MeasureTiming>>
deserialize(const std::vector<hardware::neuralnetworks::V1_2::FmqRequestDatum>& data);
/**
* RequestChannelReceiver is responsible for waiting on the channel until the
* packet is available, extracting the packet from the channel, and
* deserializing the packet.
*
* Because the receiver can wait on a packet that may never come (e.g., because
* the sending side of the packet has been closed), this object can be
* invalidated, unblocking the receiver.
*/
class RequestChannelReceiver {
using FmqRequestChannel =
hardware::MessageQueue<hardware::neuralnetworks::V1_2::FmqRequestDatum,
hardware::kSynchronizedReadWrite>;
public:
/**
* Create the receiving end of a request channel.
*
* Prefer this call over the constructor.
*
* @param requestChannel Descriptor for the request channel.
* @param pollingTimeWindow How much time (in microseconds) the
* RequestChannelReceiver is allowed to poll the FMQ before waiting on
* the blocking futex. Polling may result in lower latencies at the
* potential cost of more power usage.
* @return RequestChannelReceiver on successful creation, nullptr otherwise.
*/
static std::unique_ptr<RequestChannelReceiver> create(
const FmqRequestDescriptor& requestChannel,
std::chrono::microseconds pollingTimeWindow);
/**
* Get the request from the channel.
*
* This method will block until either:
* 1) The packet has been retrieved, or
* 2) The receiver has been invalidated
*
* @return Request object if successfully received, std::nullopt if error or
* if the receiver object was invalidated.
*/
std::optional<std::tuple<hardware::neuralnetworks::V1_0::Request, std::vector<int32_t>,
hardware::neuralnetworks::V1_2::MeasureTiming>>
getBlocking();
/**
* Method to mark the channel as invalid, unblocking any current or future
* calls to RequestChannelReceiver::getBlocking.
*/
void invalidate();
RequestChannelReceiver(std::unique_ptr<FmqRequestChannel> fmqRequestChannel,
std::chrono::microseconds pollingTimeWindow);
private:
std::optional<std::vector<hardware::neuralnetworks::V1_2::FmqRequestDatum>> getPacketBlocking();
const std::unique_ptr<FmqRequestChannel> mFmqRequestChannel;
std::atomic<bool> mTeardown{false};
const std::chrono::microseconds kPollingTimeWindow;
};
/**
* ResultChannelSender is responsible for serializing the result packet of
* information, sending it on the result channel, and signaling that the data is
* available.
*/
class ResultChannelSender {
using FmqResultChannel = hardware::MessageQueue<hardware::neuralnetworks::V1_2::FmqResultDatum,
hardware::kSynchronizedReadWrite>;
public:
/**
* Create the sending end of a result channel.
*
* Prefer this call over the constructor.
*
* @param resultChannel Descriptor for the result channel.
* @return ResultChannelSender on successful creation, nullptr otherwise.
*/
static std::unique_ptr<ResultChannelSender> create(const FmqResultDescriptor& resultChannel);
/**
* Send the result to the channel.
*
* @param errorStatus Status of the execution.
* @param outputShapes Dynamic shapes of the output tensors.
* @param timing Timing information of the execution.
* @return 'true' on successful send, 'false' otherwise.
*/
bool send(hardware::neuralnetworks::V1_0::ErrorStatus errorStatus,
const std::vector<hardware::neuralnetworks::V1_2::OutputShape>& outputShapes,
hardware::neuralnetworks::V1_2::Timing timing);
// prefer calling ResultChannelSender::send
bool sendPacket(const std::vector<hardware::neuralnetworks::V1_2::FmqResultDatum>& packet);
ResultChannelSender(std::unique_ptr<FmqResultChannel> fmqResultChannel);
private:
const std::unique_ptr<FmqResultChannel> mFmqResultChannel;
};
/**
* The ExecutionBurstServer class is responsible for waiting for and
* deserializing a request object from a FMQ, performing the inference, and
* serializing the result back across another FMQ.
*/
class ExecutionBurstServer : public hardware::neuralnetworks::V1_2::IBurstContext {
DISALLOW_IMPLICIT_CONSTRUCTORS(ExecutionBurstServer);
public:
/**
* IBurstExecutorWithCache is a callback object passed to
* ExecutionBurstServer's factory function that is used to perform an
* execution. Because some memory resources are needed across multiple
* executions, this object also contains a local cache that can directly be
* used in the execution.
*
* ExecutionBurstServer will never access its IBurstExecutorWithCache object
* with concurrent calls.
*/
class IBurstExecutorWithCache {
DISALLOW_COPY_AND_ASSIGN(IBurstExecutorWithCache);
public:
IBurstExecutorWithCache() = default;
virtual ~IBurstExecutorWithCache() = default;
/**
* Checks if a cache entry specified by a slot is present in the cache.
*
* @param slot Identifier of the cache entry.
* @return 'true' if the cache entry is present in the cache, 'false'
* otherwise.
*/
virtual bool isCacheEntryPresent(int32_t slot) const = 0;
/**
* Adds an entry specified by a slot to the cache.
*
* The caller of this function must ensure that the cache entry that is
* being added is not already present in the cache. This can be checked
* via isCacheEntryPresent.
*
* @param memory Memory resource to be cached.
* @param slot Slot identifier corresponding to the memory resource.
*/
virtual void addCacheEntry(const hardware::hidl_memory& memory, int32_t slot) = 0;
/**
* Removes an entry specified by a slot from the cache.
*
* If the cache entry corresponding to the slot number does not exist,
* the call does nothing.
*
* @param slot Slot identifier corresponding to the memory resource.
*/
virtual void removeCacheEntry(int32_t slot) = 0;
/**
* Perform an execution.
*
* @param request Request object with inputs and outputs specified.
* Request::pools is empty, and DataLocation::poolIndex instead
* refers to the 'slots' argument as if it were Request::pools.
* @param slots Slots corresponding to the cached memory entries to be
* used.
* @param measure Whether timing information is requested for the
* execution.
* @return Result of the execution, including the status of the
* execution, dynamic output shapes, and any timing information.
*/
virtual std::tuple<hardware::neuralnetworks::V1_0::ErrorStatus,
hardware::hidl_vec<hardware::neuralnetworks::V1_2::OutputShape>,
hardware::neuralnetworks::V1_2::Timing>
execute(const hardware::neuralnetworks::V1_0::Request& request,
const std::vector<int32_t>& slots,
hardware::neuralnetworks::V1_2::MeasureTiming measure) = 0;
};
/**
* Create automated context to manage FMQ-based executions.
*
* This function is intended to be used by a service to automatically:
* 1) Receive data from a provided FMQ
* 2) Execute a model with the given information
* 3) Send the result to the created FMQ
*
* @param callback Callback used to retrieve memories corresponding to
* unrecognized slots.
* @param requestChannel Input FMQ channel through which the client passes the
* request to the service.
* @param resultChannel Output FMQ channel from which the client can retrieve
* the result of the execution.
* @param executorWithCache Object which maintains a local cache of the
* memory pools and executes using the cached memory pools.
* @param pollingTimeWindow How much time (in microseconds) the
* ExecutionBurstServer is allowed to poll the FMQ before waiting on
* the blocking futex. Polling may result in lower latencies at the
* potential cost of more power usage.
* @result IBurstContext Handle to the burst context.
*/
static sp<ExecutionBurstServer> create(
const sp<hardware::neuralnetworks::V1_2::IBurstCallback>& callback,
const FmqRequestDescriptor& requestChannel, const FmqResultDescriptor& resultChannel,
std::shared_ptr<IBurstExecutorWithCache> executorWithCache,
std::chrono::microseconds pollingTimeWindow = std::chrono::microseconds{0});
/**
* Create automated context to manage FMQ-based executions.
*
* This function is intended to be used by a service to automatically:
* 1) Receive data from a provided FMQ
* 2) Execute a model with the given information
* 3) Send the result to the created FMQ
*
* @param callback Callback used to retrieve memories corresponding to
* unrecognized slots.
* @param requestChannel Input FMQ channel through which the client passes the
* request to the service.
* @param resultChannel Output FMQ channel from which the client can retrieve
* the result of the execution.
* @param preparedModel PreparedModel that the burst object was created from.
* IPreparedModel::executeSynchronously will be used to perform the
* execution.
* @param pollingTimeWindow How much time (in microseconds) the
* ExecutionBurstServer is allowed to poll the FMQ before waiting on
* the blocking futex. Polling may result in lower latencies at the
* potential cost of more power usage.
* @result IBurstContext Handle to the burst context.
*/
static sp<ExecutionBurstServer> create(
const sp<hardware::neuralnetworks::V1_2::IBurstCallback>& callback,
const FmqRequestDescriptor& requestChannel, const FmqResultDescriptor& resultChannel,
hardware::neuralnetworks::V1_2::IPreparedModel* preparedModel,
std::chrono::microseconds pollingTimeWindow = std::chrono::microseconds{0});
ExecutionBurstServer(const sp<hardware::neuralnetworks::V1_2::IBurstCallback>& callback,
std::unique_ptr<RequestChannelReceiver> requestChannel,
std::unique_ptr<ResultChannelSender> resultChannel,
std::shared_ptr<IBurstExecutorWithCache> cachedExecutor);
~ExecutionBurstServer();
// Used by the NN runtime to preemptively remove any stored memory.
hardware::Return<void> freeMemory(int32_t slot) override;
private:
// Ensures all cache entries contained in mExecutorWithCache are present in
// the cache. If they are not present, they are retrieved (via
// IBurstCallback::getMemories) and added to mExecutorWithCache.
//
// This method is locked via mMutex when it is called.
void ensureCacheEntriesArePresentLocked(const std::vector<int32_t>& slots);
// Work loop that will continue processing execution requests until the
// ExecutionBurstServer object is freed.
void task();
std::thread mWorker;
std::mutex mMutex;
std::atomic<bool> mTeardown{false};
const sp<hardware::neuralnetworks::V1_2::IBurstCallback> mCallback;
const std::unique_ptr<RequestChannelReceiver> mRequestChannelReceiver;
const std::unique_ptr<ResultChannelSender> mResultChannelSender;
const std::shared_ptr<IBurstExecutorWithCache> mExecutorWithCache;
};
} // namespace android::nn
#endif // ANDROID_FRAMEWORKS_ML_NN_COMMON_EXECUTION_BURST_SERVER_H