Files
hardware_interfaces/neuralnetworks/utils/common/include/nnapi/hal/CommonUtils.h
Xusong Wang 5f6bedb43a Introduce reusable execution to canonical interface -- HAL.
This CL modifies the canonical interface for reusable executions:
- Add new interface: IExecution with compute and computeFenced methods
- Add new method IPreparedModel::createExecution

In NNAPI runtime, the new interface IExecution is used to
memoize request-specific execution resources (e.g. converted HAL
request). The expected usage is that, IPreparedModel::createExecution
will be invoked in the first computation of a reusable NDK ANNExecution
object, and IExecution::compute* will be invoked repeatedly.

The IPreparedModel::execute* methods are preserved to avoid redundant
object creation and memoization overhead for a single-time
(non-reusable) execution.

For a vendor implementing the canonical interfaces, only the
IPreparedModel::execute* methods will be called because there is
currently no reusable execution at HAL interface. A DefaultExecution
implementation is provided to reduce the work needed on the vendor side.

Bug: 184073769
Test: NNT_static
Test: neuralnetworks_utils_hal_1_0_test
Test: neuralnetworks_utils_hal_1_1_test
Test: neuralnetworks_utils_hal_1_2_test
Test: neuralnetworks_utils_hal_1_3_test
Test: neuralnetworks_utils_hal_common_test
Test: neuralnetworks_utils_hal_aidl_test
Change-Id: I91790bb5ccf5ae648687fe603f88ffda2c9fd2b2
Merged-In: I91790bb5ccf5ae648687fe603f88ffda2c9fd2b2
(cherry picked from commit 727a7b2104)
2021-05-10 15:21:36 -07:00

143 lines
6.4 KiB
C++

/*
* Copyright (C) 2020 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_HARDWARE_INTERFACES_NEURALNETWORKS_UTILS_COMMON_COMMON_UTILS_H
#define ANDROID_HARDWARE_INTERFACES_NEURALNETWORKS_UTILS_COMMON_COMMON_UTILS_H
#include <cutils/native_handle.h>
#include <hidl/HidlSupport.h>
#include <nnapi/Result.h>
#include <nnapi/SharedMemory.h>
#include <nnapi/Types.h>
#include <functional>
#include <vector>
// Shorthands
namespace android::hardware::neuralnetworks {
namespace hal = ::android::hardware::neuralnetworks;
} // namespace android::hardware::neuralnetworks
// Shorthands
namespace aidl::android::hardware::neuralnetworks {
namespace aidl_hal = ::aidl::android::hardware::neuralnetworks;
namespace hal = ::android::hardware::neuralnetworks;
namespace nn = ::android::nn;
} // namespace aidl::android::hardware::neuralnetworks
// Shorthands
namespace android::nn {
namespace hal = ::android::hardware::neuralnetworks;
namespace aidl_hal = ::aidl::android::hardware::neuralnetworks;
} // namespace android::nn
namespace android::hardware::neuralnetworks::utils {
nn::Capabilities::OperandPerformanceTable makeQuantized8PerformanceConsistentWithP(
const nn::Capabilities::PerformanceInfo& float32Performance,
const nn::Capabilities::PerformanceInfo& quantized8Performance);
// Indicates if the object contains no pointer-based data that could be relocated to shared memory.
bool hasNoPointerData(const nn::Model& model);
bool hasNoPointerData(const nn::Request& request);
// Relocate pointer-based data to shared memory. If `model` has no Operand::LifeTime::POINTER data,
// the function returns with a reference to `model`. If `model` has Operand::LifeTime::POINTER data,
// the model is copied to `maybeModelInSharedOut` with the POINTER data relocated to a memory pool,
// and the function returns with a reference to `*maybeModelInSharedOut`.
nn::GeneralResult<std::reference_wrapper<const nn::Model>> flushDataFromPointerToShared(
const nn::Model* model, std::optional<nn::Model>* maybeModelInSharedOut);
// Record a relocation mapping between pointer-based data and shared memory.
// Only two specializations of this template may exist:
// - RelocationInfo<const void*> for request inputs
// - RelocationInfo<void*> for request outputs
template <typename PointerType>
struct RelocationInfo {
PointerType data;
size_t length;
size_t offset;
};
using InputRelocationInfo = RelocationInfo<const void*>;
using OutputRelocationInfo = RelocationInfo<void*>;
// Keep track of the relocation mapping between pointer-based data and shared memory pool,
// and provide method to copy the data between pointers and the shared memory pool.
// Only two specializations of this template may exist:
// - RelocationTracker<InputRelocationInfo> for request inputs
// - RelocationTracker<OutputRelocationInfo> for request outputs
template <typename RelocationInfoType>
class RelocationTracker {
public:
static nn::GeneralResult<std::unique_ptr<RelocationTracker>> create(
std::vector<RelocationInfoType> relocationInfos, nn::SharedMemory memory) {
auto mapping = NN_TRY(map(memory));
return std::make_unique<RelocationTracker<RelocationInfoType>>(
std::move(relocationInfos), std::move(memory), std::move(mapping));
}
RelocationTracker(std::vector<RelocationInfoType> relocationInfos, nn::SharedMemory memory,
nn::Mapping mapping)
: kRelocationInfos(std::move(relocationInfos)),
kMemory(std::move(memory)),
kMapping(std::move(mapping)) {}
// Specializations defined in CommonUtils.cpp.
// For InputRelocationTracker, this method will copy pointer data to the shared memory pool.
// For OutputRelocationTracker, this method will copy shared memory data to the pointers.
void flush() const;
private:
const std::vector<RelocationInfoType> kRelocationInfos;
const nn::SharedMemory kMemory;
const nn::Mapping kMapping;
};
using InputRelocationTracker = RelocationTracker<InputRelocationInfo>;
using OutputRelocationTracker = RelocationTracker<OutputRelocationInfo>;
struct RequestRelocation {
std::unique_ptr<InputRelocationTracker> input;
std::unique_ptr<OutputRelocationTracker> output;
};
// Relocate pointer-based data to shared memory. If `request` has no
// Request::Argument::LifeTime::POINTER data, the function returns with a reference to `request`. If
// `request` has Request::Argument::LifeTime::POINTER data, the request is copied to
// `maybeRequestInSharedOut` with the POINTER data relocated to a memory pool, and the function
// returns with a reference to `*maybeRequestInSharedOut`. The `relocationOut` will be set to track
// the input and output relocations.
//
// Unlike `flushDataFromPointerToShared`, this method will not copy the input pointer data to the
// shared memory pool. Use `relocationOut` to flush the input or output data after the call.
nn::GeneralResult<std::reference_wrapper<const nn::Request>> convertRequestFromPointerToShared(
const nn::Request* request, std::optional<nn::Request>* maybeRequestInSharedOut,
RequestRelocation* relocationOut);
nn::GeneralResult<std::vector<uint32_t>> countNumberOfConsumers(
size_t numberOfOperands, const std::vector<nn::Operation>& operations);
nn::GeneralResult<hidl_memory> createHidlMemoryFromSharedMemory(const nn::SharedMemory& memory);
nn::GeneralResult<nn::SharedMemory> createSharedMemoryFromHidlMemory(const hidl_memory& memory);
nn::GeneralResult<hidl_handle> hidlHandleFromSharedHandle(const nn::Handle& handle);
nn::GeneralResult<nn::Handle> sharedHandleFromNativeHandle(const native_handle_t* handle);
nn::GeneralResult<hidl_vec<hidl_handle>> convertSyncFences(
const std::vector<nn::SyncFence>& fences);
} // namespace android::hardware::neuralnetworks::utils
#endif // ANDROID_HARDWARE_INTERFACES_NEURALNETWORKS_UTILS_COMMON_COMMON_UTILS_H