tequilaOS/platform_hardware_interfaces
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Add IBurst to NN AIDL HAL -- hal am: 88e6e9e66f am: 68a4a14ff4 am: a2381f49cc

Browse source 
Original change: https://android-review.googlesource.com/c/platform/hardware/interfaces/+/1645422

Change-Id: Ie056e56a0629b78392265af070df5a31eb723460
This commit is contained in:
Michael Butler 2021-03-22 20:38:14 +00:00 committed by Automerger Merge Worker
commit cc0e49681c
4 changed files with 179 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

39
neuralnetworks/aidl/aidl_api/android.hardware.neuralnetworks/current/android/hardware/neuralnetworks/IBurst.aidl Normal file
View file

@ -0,0 +1,39 @@
/*
* Copyright (C) 2021 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.
*/
///////////////////////////////////////////////////////////////////////////////
// THIS FILE IS IMMUTABLE. DO NOT EDIT IN ANY CASE. //
///////////////////////////////////////////////////////////////////////////////
// This file is a snapshot of an AIDL file. Do not edit it manually. There are
// two cases:
// 1). this is a frozen version file - do not edit this in any case.
// 2). this is a 'current' file. If you make a backwards compatible change to
// the interface (from the latest frozen version), the build system will
// prompt you to update this file with `m <name>-update-api`.
//
// You must not make a backward incompatible change to any AIDL file built
// with the aidl_interface module type with versions property set. The module
// type is used to build AIDL files in a way that they can be used across
// independently updatable components of the system. If a device is shipped
// with such a backward incompatible change, it has a high risk of breaking
// later when a module using the interface is updated, e.g., Mainline modules.
package android.hardware.neuralnetworks;
@VintfStability
interface IBurst {
android.hardware.neuralnetworks.ExecutionResult executeSynchronously(in android.hardware.neuralnetworks.Request request, in long[] memoryIdentifierTokens, in boolean measureTiming, in long deadline, in long loopTimeoutDuration);
void releaseMemoryResource(in long memoryIdentifierToken);
}

1
neuralnetworks/aidl/aidl_api/android.hardware.neuralnetworks/current/android/hardware/neuralnetworks/IPreparedModel.aidl
View file

@ -36,6 +36,7 @@ package android.hardware.neuralnetworks;
interface IPreparedModel {
android.hardware.neuralnetworks.ExecutionResult executeSynchronously(in android.hardware.neuralnetworks.Request request, in boolean measureTiming, in long deadline, in long loopTimeoutDuration);
android.hardware.neuralnetworks.FencedExecutionResult executeFenced(in android.hardware.neuralnetworks.Request request, in ParcelFileDescriptor[] waitFor, in boolean measureTiming, in long deadline, in long loopTimeoutDuration, in long duration);
android.hardware.neuralnetworks.IBurst configureExecutionBurst();
const long DEFAULT_LOOP_TIMEOUT_DURATION_NS = 2000000000;
const long MAXIMUM_LOOP_TIMEOUT_DURATION_NS = 15000000000;
}

120
neuralnetworks/aidl/android/hardware/neuralnetworks/IBurst.aidl Normal file
View file

@ -0,0 +1,120 @@
/*
* Copyright (C) 2021 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.
*/
package android.hardware.neuralnetworks;
import android.hardware.neuralnetworks.ErrorStatus;
import android.hardware.neuralnetworks.ExecutionResult;
import android.hardware.neuralnetworks.Request;
/**
* IBurst represents a burst execution object.
*
* Burst executions are a sequence of executions of the same prepared model that occur in rapid
* succession, such as frames of a camera capture or successive audio samples. A burst object is
* used to control a set of burst executions, and to preserve resources between executions, enabling
* executions to have lower overhead. Burst objects enable some optimizations:
* (1) A burst object is created before a sequence of executions, and freed when the sequence has
* ended. Because of this, the lifetime of the burst object hints to a driver how long it should
* remain in a high performance state.
* (2) A burst object can preserve resources between executions. For example, a driver can map a
* memory object on the first execution and cache the mapping in the burst object for reuse in
* subsequent executions. Any cached resource can be released when the burst object is destroyed
* or when the NNAPI runtime notifies the burst object that the resource is no longer required.
* (3) A burst object may be used for at most one execution at a time. This enables any transient
* execution resources such as intermediate tensors to be allocated once when the burst object
* is created and freed when the burst object is destroyed.
*/
@VintfStability
interface IBurst {
/**
* Performs a synchronous execution on a burst object.
*
* The execution is performed synchronously with respect to the caller. executeSynchronously
* must verify the inputs to the function are correct, and the usages of memory pools allocated
* by IDevice::allocate are valid. If there is an error, executeSynchronously must immediately
* return a service specific exception with the appropriate ErrorStatus value. If the inputs to
* the function are valid and there is no error, executeSynchronously must perform the
* execution, and must not return until the execution is complete.
*
* The caller must not change the content of any data object referenced by 'request' (described
* by the {@link DataLocation} of a {@link RequestArgument}) until executeSynchronously returns.
* executeSynchronously must not change the content of any of the data objects corresponding to
* 'request' inputs.
*
* If the burst object was configured from a prepared model wherein all tensor operands have
* fully specified dimensions, and the inputs to the function are valid, and at execution time
* every operation's input operands have legal values, then the execution should complete
* successfully: there must be no failure unless the device itself is in a bad state.
*
* executeSynchronously may be called with an optional deadline. If the execution is not able to
* be completed before the provided deadline, the execution may be aborted, and either
* {@link ErrorStatus::MISSED_DEADLINE_TRANSIENT} or {@link
* ErrorStatus::MISSED_DEADLINE_PERSISTENT} may be returned. The error due to an abort must be
* sent the same way as other errors, described above.
*
* Only a single execution on a given burst object may be active at any time.
*
* @param request The input and output information on which the prepared model is to be
* executed.
* @param memoryIdentifierTokens A list of tokens where each token is a non-negative number
* that uniquely identifies a memory object. Each memory
* identifier token corresponds to an element of request.pools. A
* value of -1 indicates no identity.
* @param measure Specifies whether or not to measure duration of the execution. The duration
* runs from the time the driver sees the call to the executeSynchronously
* function to the time the driver returns from the function.
* @param deadline The time by which the execution is expected to complete. The time is measured
* in nanoseconds since epoch of the steady clock (as from
* std::chrono::steady_clock). If the execution cannot be finished by the
* deadline, the execution may be aborted. Passing -1 means the deadline is
* omitted. Other negative values are invalid.
* @param loopTimeoutDuration The maximum amount of time in nanoseconds that should be spent
* executing a {@link OperationType::WHILE} operation. If a loop
* condition model does not output false within this duration, the
* execution must be aborted. If -1 is provided, the maximum amount
* of time is {@link DEFAULT_LOOP_TIMEOUT_DURATION_NS}. Other
* negative values are invalid. When provided, the duration must not
* exceed {@link MAXIMUM_LOOP_TIMEOUT_DURATION_NS}.
* @return ExecutionResult parcelable, containing the status of the execution, output shapes and
* timing information.
* @throws ServiceSpecificException with one of the following ErrorStatus values:
* - DEVICE_UNAVAILABLE if driver is offline or busy
* - GENERAL_FAILURE if there is an unspecified error
* - INVALID_ARGUMENT if one of the input arguments is invalid
* - MISSED_DEADLINE_* if the execution is aborted because it cannot be completed by the
* deadline
* - RESOURCE_EXHAUSTED_* if the task was aborted by the driver
*/
ExecutionResult executeSynchronously(in Request request, in long[] memoryIdentifierTokens,
in boolean measureTiming, in long deadline, in long loopTimeoutDuration);
/**
* releaseMemoryResource is used by the client to signal to the service that a memory buffer
* corresponding to a slot number is no longer needed by the client, and any cached resources
* associated with that memory object may be released.
*
* The identifier tokens are unique to the burst object.
*
* @param memoryIdentifierToken Value uniquely identifying a memory object that is no longer
* used.
* @throws ServiceSpecificException with one of the following ErrorStatus values:
* - DEVICE_UNAVAILABLE if driver is offline or busy
* - GENERAL_FAILURE if there is an unspecified error
* - INVALID_ARGUMENT if one of the input arguments is invalid
*/
void releaseMemoryResource(in long memoryIdentifierToken);
}

19
neuralnetworks/aidl/android/hardware/neuralnetworks/IPreparedModel.aidl
View file

@ -20,6 +20,7 @@ import android.hardware.common.NativeHandle;
import android.hardware.neuralnetworks.ErrorStatus;
import android.hardware.neuralnetworks.ExecutionResult;
import android.hardware.neuralnetworks.FencedExecutionResult;
import android.hardware.neuralnetworks.IBurst;
import android.hardware.neuralnetworks.Request;
/**
@ -166,4 +167,22 @@ interface IPreparedModel {
FencedExecutionResult executeFenced(in Request request, in ParcelFileDescriptor[] waitFor,
in boolean measureTiming, in long deadline, in long loopTimeoutDuration,
in long duration);
/**
* Configure a Burst object used to execute multiple inferences on a prepared model in rapid
* succession.
*
* If the prepared model was prepared from a model wherein all tensor operands have fully
* specified dimensions, and a valid serialized Request is sent to the Burst for execution, and
* at execution time every operation's input operands have legal values, then the execution
* should complete successfully (ErrorStatus::NONE): There must be no failure unless the device
* itself is in a bad state.
*
* @return burst Execution burst controller object.
* @throws ServiceSpecificException with one of the following ErrorStatus values:
* - DEVICE_UNAVAILABLE if driver is offline or busy
* - GENERAL_FAILURE if there is an unspecified error
* - RESOURCE_EXHAUSTED_* if the task was aborted by the driver
*/
IBurst configureExecutionBurst();
}