Certain mutation testing -- mutateOperandLifeTimeTest and
mutateOperandInputOutputTest -- can introduce potentially very large
CONSTANT_COPY operands, which can in turn create potentially very
large Models which must be passed across binder. To avoid overflowing
the binder buffer, we estimate the size of the mutated Model, and skip
the test if that size is too high. The old logic recognizes that our
tests only have a single active binder transaction at a time, and
assumes that there are no other clients using the same service at the
same time, and so we should have the binder buffer to ourselves; to be
conservative, we reject any Model whose estimated size exceeds half
the binder buffer size. Unfortunately, sometimes the binder buffer
still overflows, because it unexpectedly contains an allocation from
some other transaction: It appears that binder buffer memory
management is not serialized with respect to transactions from our
tests, and therefore depending on scheduler behavior, there may be a
sizeable allocation still in the buffer when we attempt to pass the
large Model. To fix this problem we become even more conservative,
and instead of limiting the Model to half the binder buffer size, we
limit it to half IBinder.MAX_IPC_SIZE (the recommended transaction
size limit). To confirm that this change does not exclude too many
tests, I checked how may times the size filter function
exceedsBinderSizeLimit is called, how many times it rejects a model
under the new logic (modelsExceedHalfMaxIPCSize), and how many times
it rejects a model under the old logic (modelsExceedHalfMaxIPCSize).
Test: VtsHalNeuralnetworksV1_0TargetTest --gtest_filter=TestGenerated/ValidationTest.Test/*-*dsp*
Test: # models = 3592, modelsExceedHalfMaxIPCSize = 212, modelsExceedHalfBufferSize = 18
Test: VtsHalNeuralnetworksV1_1TargetTest --gtest_filter=TestGenerated/ValidationTest.Test/*-*dsp*
Test: # models = 7228, modelsExceedHalfMaxIPCSize = 330, modelsExceedHalfBufferSize = 28
Test: VtsHalNeuralnetworksV1_2TargetTest --gtest_filter=TestGenerated/ValidationTest.Test/*-*dsp*
Test: # models = 52072, modelsExceedHalfMaxIPCSize = 506, modelsExceedHalfBufferSize = 28
Test: VtsHalNeuralnetworksV1_3TargetTest --gtest_filter=TestGenerated/ValidationTest.Test/*-*dsp*
Test: # models = 73342, modelsExceedHalfMaxIPCSize = 568, modelsExceedHalfBufferSize = 28
Test: VtsHalNeuralnetworksTargetTest
Bug: 227719657
Bug: 227719752
Bug: 231928847
Bug: 238777741
Bug: 242271308
Change-Id: I3f81d71ca3c0ad4c639096b1dc034a8909bc8971
Also skip FL6 (AIDL_V2) tests for older AIDL drivers.
Cherrypicked from I689fef0945428f6548977628e3c43628dd1e5bf7
Bug: 206089870
Test: VtsHalNeuralnetworksTargetTest
Specifically, for old driver such as AIDL_V1 sample driver, it can pass
HIDL tests and skip AIDL_V2 tests. For new driver such as AIDL_V2
sample driver, it can pass all tests.
Change-Id: I689fef0945428f6548977628e3c43628dd1e5bf7
(cherry picked from commit 23d4e5e298)
- Add BATCH_MATMUL operation
- Support TENSOR_INT32 for RESHAPE operation.
Also update "current" version snapshot and use
android.hardware.neuralnetworks-V2-ndk since AIDL v1 has been frozen.
Cherrypicked from Iabe45c57e2306d61055f711eda03b80b9cbe906d
Bug: 206089870
Test: mm
Change-Id: Iabe45c57e2306d61055f711eda03b80b9cbe906d
Merged-In: Iabe45c57e2306d61055f711eda03b80b9cbe906d
(cherry picked from commit aaeda0e84f)
$ m android.hardware.neuralnetworks-update-api
$ m android.hardware.neuralnetworks-freeze-api
(This cherry pick required hand editing to ensure makefile
references are changed from
android.hardware.neuralnetworks-V2-ndk_platform to
android.hardware.neuralnetworks-V3-ndk_platform.)
Bug: 202280925
Test: adb shell NeuralNetworksTest_static
Test: atest VtsHalNeuralnetworksTargetTest
Merged-In: I2b2755d7376bb847b15b395e280bf352b5b9ef55
Change-Id: I2b2755d7376bb847b15b395e280bf352b5b9ef55
(cherry picked from commit 68e4300c72)
Adds operations MIRROR_PAD and REVERSE.
Extends RSQRT to support QUANT8_ASYMM and QUANT8_ASYMM_SIGNED.
DOES NOT include tests or CPU reference implementation.
(This is a slightly non-trivial cherry pick that had to be
hand-edited to compensate for the fact that certain nn::Version
changes and makefile restructurings are absent from the target branch.)
Bug: 202280917
Test: NeuralNetworksTest_static
Test: VtsHalNeuralnetworksTargetTest
Merged-In: I0b2133346b996849faac00c46885e3633c78f024
Change-Id: I0b2133346b996849faac00c46885e3633c78f024
(cherry picked from commit 04ed8595b5)
Also skip FL6 (AIDL_V2) tests for older AIDL drivers.
Cherrypicked from I689fef0945428f6548977628e3c43628dd1e5bf7
Bug: 206089870
Test: VtsHalNeuralnetworksTargetTest
Specifically, for old driver such as AIDL_V1 sample driver, it can pass
HIDL tests and skip AIDL_V2 tests. For new driver such as AIDL_V2
sample driver, it can pass all tests.
Change-Id: I689fef0945428f6548977628e3c43628dd1e5bf7
- Add BATCH_MATMUL operation
- Support TENSOR_INT32 for RESHAPE operation.
Also update "current" version snapshot and use
android.hardware.neuralnetworks-V2-ndk since AIDL v1 has been frozen.
Cherrypicked from Iabe45c57e2306d61055f711eda03b80b9cbe906d
Bug: 206089870
Test: mm
Change-Id: Iabe45c57e2306d61055f711eda03b80b9cbe906d
Merged-In: Iabe45c57e2306d61055f711eda03b80b9cbe906d
Prior to this CL, the AHardwareBuffer in TestBlobAHWB is released in the
destructor, but later used (unlock) during the destruction of the
mMapping member. This CL fixed this issue by managing the lifetime of
AHardwareBuffer with SharedMemory.
Bug: 197199690
Test: VtsHalNeuralnetworksTargetTest
Change-Id: I00748aaaa1a3a3d9b3b62bedb77a655ddb6e210f
Merged-In: I00748aaaa1a3a3d9b3b62bedb77a655ddb6e210f
(cherry picked from commit d2ecde5c54)
This change adds the AndroidTest.xml flag --gtest_break_on_failure to
cause the gtest to terminate after an error has been reached. This early
termination is important in the case where an NN HAL service crashes
mid-test, and all remaining tests would otherwise continue to run.
Bug: 197035200
Test: m vts -j
Test: vts-tradefed
Change-Id: I0b9a14345475e432b93f92c23010a8b39712443a
Merged-In: I0b9a14345475e432b93f92c23010a8b39712443a
(cherry picked from commit ac45a5d77e)
This CL adds a check during SetUp that an NNAPI driver service is still
alive by pinging the driver service. If it is not alive, the test will
fail during SetUp. Without this quick-fail, the test case would continue
as if the driver were still active, which would result in multiple
EXPECT_* and ASSERT_* statements failing instead of a single, clear
failure message.
Bug: 197035200
Test: mma
Test: presubmit: VtsHalNeuralnetworks*TargetTest
Change-Id: Ib1b75ed20f764055699590581d5ad4e5aff4baae
Merged-In: Ib1b75ed20f764055699590581d5ad4e5aff4baae
(cherry picked from commit 9c3c8642fb)
Data and model numbers were switched in the AIDL implementation of
canonical Device.
Bug: 190757709
Test: neuralnetworks_utils_hal_aidl_test
Change-Id: I0d95b2d436994ffc877a4e02eb31f449b983e61e
This CL modifies the AIDL generated tests to use readonly memory pool
for request inputs.
Bug: 188104713
Test: VtsHalNeuralnetworksTargetTest
Change-Id: I9e62f1cdeb501bf29bcb9c56317a452c9105b272
This CL modifies the shared memory allocation for pointer arguments to
use proper alignment and padding. We use default alignment (64) and
min padding (1) for HIDL drivers, and default alignment (64) and default
padding (64) for sAIDL drivers.
Bug: 184164929
Test: NNT_static
Change-Id: I22591640fa047d5f75d437edac1a7645d3b05526
This CL modifies the canonical interface for reusable burst executions:
- Add new method IBurst::createExecution
The reusable burst execution will not fallback to another execution path
if sending request packet fails. The behavior of single-time burst
execution remains unchanged.
Additionally, this CL enables pointer -> shared memory conversion in
1.2/1.3 burst implementation.
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
Change-Id: Iaac81668d247c2cb76d70e6abbd10f00b397b19f
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
This CL alters the AIDL utility "getDevices" to return whether or not a
service is updatable. This CL additional removes IDevice::isUpdatable
because it is now redundant with the changes to "getDevices".
Bug: 170696350
Test: mma
Test: NeuralNetworksTest_static
Change-Id: I86228763937c0dd2ee5b1d9d547c367cd7504ed4
The change adds "Ns" suffix to timeOnDevice, timeInDriver,
loopTimeoutDuration, deadline and duration.
Fix: 183118329
Test: mm
Change-Id: I4afe34f27fc8717ac28c8fa09edda9918641c39a
Merged-In: Id1f9ee4b8e41873c97690bb19a5e84572dd9ccf1
The change adds "Ns" suffix to timeOnDevice, timeInDriver,
loopTimeoutDuration, deadline and duration.
Fix: 183118329
Test: mm
Change-Id: Id1f9ee4b8e41873c97690bb19a5e84572dd9ccf1
This CL enables VtsHalNeuralnetworksTargetTest to use lazy services by
changing from AServiceManager_getService (which will return nullptr for
the service if the service is not able to be loaded within a short
amount of time) to AServiceManager_waitForService (which will wait for a
longer time, allowing lazy services to start up).
Similarly, the utility code is changed from using
AServiceManager_getService to AServiceManager_waitForService where
possible.
This CL also introduces an "InvalidDevice" utility class to the
nnapi/hal/aidl utility code. InvalidDevices are minimal devices that
support no functionality but are still able to pass VTS tests.
Bug: 170696939
Test: mma
Test: VtsHalNeuralnetworksTargetTest
Change-Id: I4f806b104ef6af863ec55c2c3f2a2dd1f72b9633
This CL fixes the compiler errors that arose of changing the Memory
representation of the NN AIDL HAL, and updates the conversion and
utility code to work with the new Memory type.
This change also makes libaidlcommonsupport available to apex modules at
min sdk level 29.
Bug: 183118727
Test: mma
Test: VtsHalNeuralnetworksTargetTest
Change-Id: Ief565473b4d82e0bb43785fc3b8275b16bd26cf6
Merged-In: Ief565473b4d82e0bb43785fc3b8275b16bd26cf6
(cherry picked from commit b0fcb3927d)
Prior to this change, the NN AIDL HAL created a Memory struct analogous
to hidl_memory, consisting of a name, size, and native handle. However,
this Memory struct is not very structured, and requires both the client
and server to agree on how the data should be interpreted.
This CL tightens the structure of the Memory representation by
introducing Ashmem and MappableFile structs to android.hardware.common
in order to hold specific file descriptors representing memory regions.
Further, this CL redefines android.hardwre.neuralnetworks's Memory
object as a union of the Ashmem, MappableFile, and (existing)
HardwareBuffer memory types.
This change also adds "com.android.neuralnetworks" to the graphics
AIDL HAL's apex_available build rule.
Bug: 183118727
Test: mma
Change-Id: I32322df676b83597c9e95f13662c322a6a80accc
Merged-In: I32322df676b83597c9e95f13662c322a6a80accc
(cherry picked from commit 1158c80ff6)
Previously, the NNAPI used std::chrono::steady_clock to represent and
measure timings. However, steady_clock does not count while the system
is suspended. Instead, boot_clock is monotonic like steady_clock but
does include the time when the system is suspended.
This change also indicates that services may convert from
std::chrono::steady_clock::time_point to
android::base::boot_clock::time_point in the HIDL 1.3 NN HAL.
Bug: 183118340
Test: mma
Test: VtsHalNeuralnetworksV1_3TargetTest
Test: VtsHalNeuralnetworksTargetTest
Test: presubmit
Change-Id: I5a7d039a31d9ce98602a301387ec99635f279f42
Merged-In: I5a7d039a31d9ce98602a301387ec99635f279f42
(cherry picked from commit b8cf54cf5a)
Adds deadline and loopTimeoutDuration to nn::IBurst::execute.
This CL additionally adds a test case for
IPreparedModel::configureExecutionBurst and adds the suffix "_H" to
the header guards that were missing "_H".
Bug: 180492058
Bug: 177267324
Test: mma
Test: NeuralNetworksTest_static
Change-Id: Ic9ba2fb9b4dee4e40d99ae91bb9555a58d0508d2
Merged-In: Ic9ba2fb9b4dee4e40d99ae91bb9555a58d0508d2
(cherry picked from commit bbbdb0d5f1)
"Frenquency" often refers to the number of occurrences over a period of
time, while "probability" refers to the number of occurrences of one
event over the number of occurrences of all events. "Probability" is a
better name for this field.
Fixes: 183117895
Test: VtsHalNeuralnetworksTargetTest
Test: NNT_static
Change-Id: Ic86f73b8be2aed567ae4ca17bdb3a57c658fb349
Merged-In: Ic86f73b8be2aed567ae4ca17bdb3a57c658fb349
(cherry picked from commit 46bf892f46)
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.
This CL has the following changes:
1) Adds an IBurst interface to the NN AIDL HAL
2) Adds IPreparedModel::configureExecutionBurst to create an IBurst
object
3) Extends VTS's InvalidPreparedModel to implement
configureExecutionBurst.
Bug: 180492058
Bug: 177267324
Test: mma
Test: VTS+presubmit
Change-Id: I90cb18dcc392f07f342604a7fb2edae041848251
Merged-In: I90cb18dcc392f07f342604a7fb2edae041848251
(cherry picked from commit 5dc72d5485)
There were some comment changes made to tha auto-generated AIDL files.
Bug: N/A
Test: mma
Change-Id: I1a360115b82b0c222666c5d9189a220e74390654
Merged-In: I1a360115b82b0c222666c5d9189a220e74390654
(cherry picked from commit 402497e049)
